{
  "id": 575243,
  "title": "is classification + image2image? individual model or single model for 10 datasets?",
  "url": "/competitions/waveform-inversion/discussion/575243",
  "author_name": "SeshuRaju 🧘‍♂️",
  "post_date": "2025-04-27T06:46:45.120000",
  "votes": 18,
  "comment_count": 20,
  "views": 0,
  "content": "<h1>Average of all best individual models -&gt; <strong>MAE - 64.60 for DL4SI models</strong></h1>\n<blockquote>\n  <table>\n  <thead>\n  <tr>\n  <th>Model</th>\n  <th>Source</th>\n  <th>Target</th>\n  <th>MAE</th>\n  </tr>\n  </thead>\n  <tbody>\n  <tr>\n  <td>UNetInverseModel 33M Latent32</td>\n  <td>flatvel-a</td>\n  <td>flatvel-a</td>\n  <td>6.63</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M Latent64</td>\n  <td>flatfault-a</td>\n  <td>flatfault-a</td>\n  <td>12.88</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M Latent64</td>\n  <td>curvefault-a</td>\n  <td>curvefault-a</td>\n  <td>22.44</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M Latent64</td>\n  <td>flatvel-b</td>\n  <td>flatvel-b</td>\n  <td>26.37</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M Latent64</td>\n  <td>curvevel-a</td>\n  <td>curvevel-a</td>\n  <td>46.00</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M</td>\n  <td>style-a</td>\n  <td>style-a</td>\n  <td>67.00</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M</td>\n  <td>style-b</td>\n  <td>style-b</td>\n  <td>71.41</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M</td>\n  <td>flatfault-b</td>\n  <td>flatfault-b</td>\n  <td>92.92</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M</td>\n  <td>curvevel-b</td>\n  <td>curvevel-b</td>\n  <td>123.56</td>\n  </tr>\n  <tr>\n  <td>IUnetInverseModel</td>\n  <td>curvefault-b</td>\n  <td>curvefault-b</td>\n  <td>176.84</td>\n  </tr>\n  </tbody>\n  </table>\n</blockquote>\n<hr>\n<table>\n<thead>\n<tr>\n<th>Target</th>\n<th>count</th>\n<th>mean</th>\n<th>std</th>\n<th>min</th>\n<th>25%</th>\n<th>50%</th>\n<th>75%</th>\n<th>max</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>flatvel-a</td>\n<td>212</td>\n<td>115.487</td>\n<td>94.1268</td>\n<td>6.63</td>\n<td>51.4</td>\n<td>95.255</td>\n<td>144.572</td>\n<td>502.67</td>\n</tr>\n<tr>\n<td>flatfault-a</td>\n<td>212</td>\n<td>151.515</td>\n<td>112.252</td>\n<td>12.88</td>\n<td>75.7225</td>\n<td>116.315</td>\n<td>191.968</td>\n<td>635.85</td>\n</tr>\n<tr>\n<td>curvefault-a</td>\n<td>212</td>\n<td>242.581</td>\n<td>180.301</td>\n<td>22.44</td>\n<td>135.422</td>\n<td>176.475</td>\n<td>357.52</td>\n<td>1101.14</td>\n</tr>\n<tr>\n<td>curvevel-a</td>\n<td>212</td>\n<td>244.101</td>\n<td>140.08</td>\n<td>46</td>\n<td>162.987</td>\n<td>223.74</td>\n<td>297.4</td>\n<td>889.18</td>\n</tr>\n<tr>\n<td>style-b</td>\n<td>212</td>\n<td>269.586</td>\n<td>138.38</td>\n<td>71.41</td>\n<td>213.805</td>\n<td>239.845</td>\n<td>284.812</td>\n<td>1012.49</td>\n</tr>\n<tr>\n<td>style-a</td>\n<td>212</td>\n<td>347.427</td>\n<td>134.048</td>\n<td>67</td>\n<td>281.2</td>\n<td>330.29</td>\n<td>376.275</td>\n<td>894.53</td>\n</tr>\n<tr>\n<td>flatfault-b</td>\n<td>212</td>\n<td>361.168</td>\n<td>161.03</td>\n<td>92.92</td>\n<td>314.885</td>\n<td>354.405</td>\n<td>397.238</td>\n<td>1121.32</td>\n</tr>\n<tr>\n<td>curvefault-b</td>\n<td>212</td>\n<td>450.822</td>\n<td>182.416</td>\n<td>176.84</td>\n<td>396.375</td>\n<td>430.89</td>\n<td>464.228</td>\n<td>1345.67</td>\n</tr>\n<tr>\n<td>flatvel-b</td>\n<td>212</td>\n<td>578.099</td>\n<td>247.847</td>\n<td>26.37</td>\n<td>577.227</td>\n<td>649.205</td>\n<td>705.645</td>\n<td>1056.44</td>\n</tr>\n<tr>\n<td>curvevel-b</td>\n<td>212</td>\n<td>645.475</td>\n<td>174.971</td>\n<td>123.56</td>\n<td>623.84</td>\n<td>670.83</td>\n<td>729.697</td>\n<td>1015.51</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<h1><a href=\"https://github.com/KGML-lab/Generalized-Forward-Inverse-Framework-for-DL4SI\" target=\"_blank\">Unified Framework for Forward and Inverse Problems</a></h1>\n<p><a href=\"https://github.com/KGML-lab/Generalized-Forward-Inverse-Framework-for-DL4SI/tree/main/evaluate/Metrics_final\" target=\"_blank\">Source</a><br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2Feb0f8613ecf69a7319e3b6af4c85a8af%2FScreenshot%202025-04-27%20at%2012.50.50AM.png?generation=1745736307342753&amp;alt=media\" alt=\"\"></p>\n<h2>FLATVEL-A</h2>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>6.63</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64 No Skip</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>7.45</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>7.49</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32 No Skip</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>8.29</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32 No Skip</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>8.58</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>8.66</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>9.01</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>9.77</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>11.05</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>11.09</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent8</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>11.44</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>12.11</td>\n</tr>\n<tr>\n<td>AutoLinear Inversion</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>12.16</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16 No Skip</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>13.88</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>15.74</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## FLATFAULT-A </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>12.88</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>14.52</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>15.41</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>15.53</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>15.85</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32 No Skip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>16.11</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64 No Skip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>17.04</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32 No Skip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>17.65</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64 No Skip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>19.19</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16 No Skip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>20.54</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent8</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>20.84</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>21.7</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>23.18</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>23.78</td>\n</tr>\n<tr>\n<td>AutoLinear Inversion</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>24.56</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## CURVEFAULT-A </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>22.44</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>23.4</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64 No Skip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>26.42</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>26.98</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>27.05</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>27.05</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>28</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32 No Skip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>30.12</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>32.53</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>33.53</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16 No Skip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>34.29</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>35.3</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>37.8</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32 No Skip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>38.49</td>\n</tr>\n<tr>\n<td>Velocity GAN</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>38.8</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## FLATVEL-B </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>26.37</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>26.72</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>27.1</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>27.95</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>28.12</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>29</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>29.38</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>31.42</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>33.16</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>33.76</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>34.95</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>38.65</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>40.13</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>42.97</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent8</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>43.39</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## CURVEVEL-A </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>46</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>48.06</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>49.63</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>53.19</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64 No Skip</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>54.42</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>55.61</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>56.57</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>57.87</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>59.94</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64 No Skip</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>60.99</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>62.15</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>62.92</td>\n</tr>\n<tr>\n<td>IUnetInverseModel</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>62.92</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>63.04</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32 No Skip</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>63.47</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## STYLE-A </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>67</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>67.09</td>\n</tr>\n<tr>\n<td>IUnetInverseModel</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>74.07</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>81.71</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>84.35</td>\n</tr>\n<tr>\n<td>Velocity GAN</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>91.77</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>93.96</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>97.98</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>103.36</td>\n</tr>\n<tr>\n<td>AutoLinear Inversion</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>107.85</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>style-b</td>\n<td>style-a</td>\n<td>186.53</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-b</td>\n<td>style-a</td>\n<td>187.93</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>style-b</td>\n<td>style-a</td>\n<td>192.72</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>style-b</td>\n<td>style-a</td>\n<td>194.95</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>style-b</td>\n<td>style-a</td>\n<td>196.8</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## STYLE-B </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>71.41</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>83.62</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>84.81</td>\n</tr>\n<tr>\n<td>AutoLinear Inversion</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>95.63</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>103.38</td>\n</tr>\n<tr>\n<td>Velocity GAN</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>104.52</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>106.31</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>110.17</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>112.08</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>120.73</td>\n</tr>\n<tr>\n<td>IUnetInverseModel</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>120.96</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>121.43</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>122.41</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>129.17</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>132.56</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## FLATFAULT-B </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>92.92</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>94.12</td>\n</tr>\n<tr>\n<td>IUnetInverseModel</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>95.2</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>95.24</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>95.28</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>99.48</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>113.15</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>117.12</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>120.08</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>122.25</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>125.66</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>130.89</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>138.09</td>\n</tr>\n<tr>\n<td>Velocity GAN</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>138.89</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>142.66</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## CURVEVEL-B </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>123.56</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>126.07</td>\n</tr>\n<tr>\n<td>IUnetInverseModel</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>135.35</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>139.02</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>139.82</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>144.56</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>147.12</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>148.4</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>154.09</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>156.69</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>157.16</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>166.42</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>171.53</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>174.22</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>182.81</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## CURVEFAULT-B </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>IUnetInverseModel</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>176.84</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>177.19</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>181.03</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>185.23</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>188.04</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>202.36</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>205.03</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>209.92</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>218.11</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>223.76</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>223.81</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>223.86</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>228.18</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>229.39</td>\n</tr>\n<tr>\n<td>Velocity GAN</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>235.7</td>\n</tr>\n</tbody>\n</table>",
  "messages": [
    {
      "id": 3188167,
      "postDate": "2025-04-27T06:46:45.120Z",
      "content": "<h1>Average of all best individual models -&gt; <strong>MAE - 64.60 for DL4SI models</strong></h1>\n<blockquote>\n  <table>\n  <thead>\n  <tr>\n  <th>Model</th>\n  <th>Source</th>\n  <th>Target</th>\n  <th>MAE</th>\n  </tr>\n  </thead>\n  <tbody>\n  <tr>\n  <td>UNetInverseModel 33M Latent32</td>\n  <td>flatvel-a</td>\n  <td>flatvel-a</td>\n  <td>6.63</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M Latent64</td>\n  <td>flatfault-a</td>\n  <td>flatfault-a</td>\n  <td>12.88</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M Latent64</td>\n  <td>curvefault-a</td>\n  <td>curvefault-a</td>\n  <td>22.44</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M Latent64</td>\n  <td>flatvel-b</td>\n  <td>flatvel-b</td>\n  <td>26.37</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M Latent64</td>\n  <td>curvevel-a</td>\n  <td>curvevel-a</td>\n  <td>46.00</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M</td>\n  <td>style-a</td>\n  <td>style-a</td>\n  <td>67.00</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M</td>\n  <td>style-b</td>\n  <td>style-b</td>\n  <td>71.41</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M</td>\n  <td>flatfault-b</td>\n  <td>flatfault-b</td>\n  <td>92.92</td>\n  </tr>\n  <tr>\n  <td>UNetInverseModel 33M</td>\n  <td>curvevel-b</td>\n  <td>curvevel-b</td>\n  <td>123.56</td>\n  </tr>\n  <tr>\n  <td>IUnetInverseModel</td>\n  <td>curvefault-b</td>\n  <td>curvefault-b</td>\n  <td>176.84</td>\n  </tr>\n  </tbody>\n  </table>\n</blockquote>\n<hr>\n<table>\n<thead>\n<tr>\n<th>Target</th>\n<th>count</th>\n<th>mean</th>\n<th>std</th>\n<th>min</th>\n<th>25%</th>\n<th>50%</th>\n<th>75%</th>\n<th>max</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>flatvel-a</td>\n<td>212</td>\n<td>115.487</td>\n<td>94.1268</td>\n<td>6.63</td>\n<td>51.4</td>\n<td>95.255</td>\n<td>144.572</td>\n<td>502.67</td>\n</tr>\n<tr>\n<td>flatfault-a</td>\n<td>212</td>\n<td>151.515</td>\n<td>112.252</td>\n<td>12.88</td>\n<td>75.7225</td>\n<td>116.315</td>\n<td>191.968</td>\n<td>635.85</td>\n</tr>\n<tr>\n<td>curvefault-a</td>\n<td>212</td>\n<td>242.581</td>\n<td>180.301</td>\n<td>22.44</td>\n<td>135.422</td>\n<td>176.475</td>\n<td>357.52</td>\n<td>1101.14</td>\n</tr>\n<tr>\n<td>curvevel-a</td>\n<td>212</td>\n<td>244.101</td>\n<td>140.08</td>\n<td>46</td>\n<td>162.987</td>\n<td>223.74</td>\n<td>297.4</td>\n<td>889.18</td>\n</tr>\n<tr>\n<td>style-b</td>\n<td>212</td>\n<td>269.586</td>\n<td>138.38</td>\n<td>71.41</td>\n<td>213.805</td>\n<td>239.845</td>\n<td>284.812</td>\n<td>1012.49</td>\n</tr>\n<tr>\n<td>style-a</td>\n<td>212</td>\n<td>347.427</td>\n<td>134.048</td>\n<td>67</td>\n<td>281.2</td>\n<td>330.29</td>\n<td>376.275</td>\n<td>894.53</td>\n</tr>\n<tr>\n<td>flatfault-b</td>\n<td>212</td>\n<td>361.168</td>\n<td>161.03</td>\n<td>92.92</td>\n<td>314.885</td>\n<td>354.405</td>\n<td>397.238</td>\n<td>1121.32</td>\n</tr>\n<tr>\n<td>curvefault-b</td>\n<td>212</td>\n<td>450.822</td>\n<td>182.416</td>\n<td>176.84</td>\n<td>396.375</td>\n<td>430.89</td>\n<td>464.228</td>\n<td>1345.67</td>\n</tr>\n<tr>\n<td>flatvel-b</td>\n<td>212</td>\n<td>578.099</td>\n<td>247.847</td>\n<td>26.37</td>\n<td>577.227</td>\n<td>649.205</td>\n<td>705.645</td>\n<td>1056.44</td>\n</tr>\n<tr>\n<td>curvevel-b</td>\n<td>212</td>\n<td>645.475</td>\n<td>174.971</td>\n<td>123.56</td>\n<td>623.84</td>\n<td>670.83</td>\n<td>729.697</td>\n<td>1015.51</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<h1><a href=\"https://github.com/KGML-lab/Generalized-Forward-Inverse-Framework-for-DL4SI\" target=\"_blank\">Unified Framework for Forward and Inverse Problems</a></h1>\n<p><a href=\"https://github.com/KGML-lab/Generalized-Forward-Inverse-Framework-for-DL4SI/tree/main/evaluate/Metrics_final\" target=\"_blank\">Source</a><br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2Feb0f8613ecf69a7319e3b6af4c85a8af%2FScreenshot%202025-04-27%20at%2012.50.50AM.png?generation=1745736307342753&amp;alt=media\" alt=\"\"></p>\n<h2>FLATVEL-A</h2>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>6.63</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64 No Skip</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>7.45</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>7.49</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32 No Skip</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>8.29</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32 No Skip</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>8.58</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>8.66</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>9.01</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>9.77</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>11.05</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>11.09</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent8</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>11.44</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>12.11</td>\n</tr>\n<tr>\n<td>AutoLinear Inversion</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>12.16</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16 No Skip</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>13.88</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>flatvel-a</td>\n<td>flatvel-a</td>\n<td>15.74</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## FLATFAULT-A </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>12.88</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>14.52</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>15.41</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>15.53</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>15.85</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32 No Skip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>16.11</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64 No Skip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>17.04</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32 No Skip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>17.65</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64 No Skip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>19.19</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16 No Skip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>20.54</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent8</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>20.84</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>21.7</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>23.18</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>23.78</td>\n</tr>\n<tr>\n<td>AutoLinear Inversion</td>\n<td>flatfault-a</td>\n<td>flatfault-a</td>\n<td>24.56</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## CURVEFAULT-A </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>22.44</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>23.4</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64 No Skip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>26.42</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>26.98</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>27.05</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>27.05</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>28</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32 No Skip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>30.12</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>32.53</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>33.53</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16 No Skip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>34.29</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>35.3</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>37.8</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32 No Skip</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>38.49</td>\n</tr>\n<tr>\n<td>Velocity GAN</td>\n<td>curvefault-a</td>\n<td>curvefault-a</td>\n<td>38.8</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## FLATVEL-B </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>26.37</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>26.72</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>27.1</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>27.95</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>28.12</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>29</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>29.38</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>31.42</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>33.16</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>33.76</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>34.95</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>38.65</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>40.13</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>42.97</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent8</td>\n<td>flatvel-b</td>\n<td>flatvel-b</td>\n<td>43.39</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## CURVEVEL-A </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>46</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>48.06</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>49.63</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>53.19</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64 No Skip</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>54.42</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>55.61</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>56.57</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>57.87</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>59.94</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64 No Skip</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>60.99</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>62.15</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>62.92</td>\n</tr>\n<tr>\n<td>IUnetInverseModel</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>62.92</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>63.04</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32 No Skip</td>\n<td>curvevel-a</td>\n<td>curvevel-a</td>\n<td>63.47</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## STYLE-A </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>67</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>67.09</td>\n</tr>\n<tr>\n<td>IUnetInverseModel</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>74.07</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>81.71</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>84.35</td>\n</tr>\n<tr>\n<td>Velocity GAN</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>91.77</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>93.96</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>97.98</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>103.36</td>\n</tr>\n<tr>\n<td>AutoLinear Inversion</td>\n<td>style-a</td>\n<td>style-a</td>\n<td>107.85</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>style-b</td>\n<td>style-a</td>\n<td>186.53</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-b</td>\n<td>style-a</td>\n<td>187.93</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>style-b</td>\n<td>style-a</td>\n<td>192.72</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>style-b</td>\n<td>style-a</td>\n<td>194.95</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>style-b</td>\n<td>style-a</td>\n<td>196.8</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## STYLE-B </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>71.41</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>83.62</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>84.81</td>\n</tr>\n<tr>\n<td>AutoLinear Inversion</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>95.63</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>103.38</td>\n</tr>\n<tr>\n<td>Velocity GAN</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>104.52</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>106.31</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>110.17</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>112.08</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>120.73</td>\n</tr>\n<tr>\n<td>IUnetInverseModel</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>120.96</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>121.43</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>122.41</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>style-b</td>\n<td>style-b</td>\n<td>129.17</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>style-a</td>\n<td>style-b</td>\n<td>132.56</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## FLATFAULT-B </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>92.92</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>94.12</td>\n</tr>\n<tr>\n<td>IUnetInverseModel</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>95.2</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>95.24</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>95.28</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>99.48</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>113.15</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>117.12</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>120.08</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>122.25</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>125.66</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>130.89</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>138.09</td>\n</tr>\n<tr>\n<td>Velocity GAN</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>138.89</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>flatfault-b</td>\n<td>flatfault-b</td>\n<td>142.66</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## CURVEVEL-B </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>123.56</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>126.07</td>\n</tr>\n<tr>\n<td>IUnetInverseModel</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>135.35</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>139.02</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>139.82</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>144.56</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>147.12</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>148.4</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>154.09</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>156.69</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>157.16</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>166.42</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>171.53</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent16</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>174.22</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>curvevel-b</td>\n<td>curvevel-b</td>\n<td>182.81</td>\n</tr>\n</tbody>\n</table>\n<hr>\n<p>## CURVEFAULT-B </p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>Source</th>\n<th>Target</th>\n<th>MAE</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>IUnetInverseModel</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>176.84</td>\n</tr>\n<tr>\n<td>Invertible XNet cycle warmup</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>177.19</td>\n</tr>\n<tr>\n<td>Invertible XNet</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>181.03</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>185.23</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent64</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>188.04</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M NoSkip</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>202.36</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent32</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>205.03</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent64</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>209.92</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M Latent32</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>218.11</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>223.76</td>\n</tr>\n<tr>\n<td>UNetInverseModel 17M NoSkip</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>223.81</td>\n</tr>\n<tr>\n<td>Invertible XNet Adam</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>223.86</td>\n</tr>\n<tr>\n<td>InversionNet</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>228.18</td>\n</tr>\n<tr>\n<td>UNetInverseModel 33M Latent16</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>229.39</td>\n</tr>\n<tr>\n<td>Velocity GAN</td>\n<td>curvefault-b</td>\n<td>curvefault-b</td>\n<td>235.7</td>\n</tr>\n</tbody>\n</table>",
      "rawMarkdown": "# Average of all best individual models -> **MAE - 64.60 for DL4SI models**\n>| Model                         | Source       | Target       |    MAE |\n|:------------------------------|:-------------|:-------------|-------:|\n| UNetInverseModel 33M Latent32 | flatvel-a    | flatvel-a    |   6.63 |\n| UNetInverseModel 33M Latent64 | flatfault-a  | flatfault-a  |  12.88 |\n| UNetInverseModel 33M Latent64 | curvefault-a | curvefault-a |  22.44 |\n| UNetInverseModel 33M Latent64 | flatvel-b    | flatvel-b    |  26.37 |\n| UNetInverseModel 33M Latent64 | curvevel-a   | curvevel-a   |  46.00    |\n| UNetInverseModel 33M          | style-a      | style-a      |  67.00    |\n| UNetInverseModel 33M          | style-b      | style-b      |  71.41 |\n| UNetInverseModel 33M          | flatfault-b  | flatfault-b  |  92.92 |\n| UNetInverseModel 33M          | curvevel-b   | curvevel-b   | 123.56 |\n| IUnetInverseModel             | curvefault-b | curvefault-b | 176.84 |\n\n---\n\n| Target       |   count |    mean |      std |    min |      25% |     50% |     75% |     max |\n|:-------------|--------:|--------:|---------:|-------:|---------:|--------:|--------:|--------:|\n| flatvel-a    |     212 | 115.487 |  94.1268 |   6.63 |  51.4    |  95.255 | 144.572 |  502.67 |\n| flatfault-a  |     212 | 151.515 | 112.252  |  12.88 |  75.7225 | 116.315 | 191.968 |  635.85 |\n| curvefault-a |     212 | 242.581 | 180.301  |  22.44 | 135.422  | 176.475 | 357.52  | 1101.14 |\n| curvevel-a   |     212 | 244.101 | 140.08   |  46    | 162.987  | 223.74  | 297.4   |  889.18 |\n| style-b      |     212 | 269.586 | 138.38   |  71.41 | 213.805  | 239.845 | 284.812 | 1012.49 |\n| style-a      |     212 | 347.427 | 134.048  |  67    | 281.2    | 330.29  | 376.275 |  894.53 |\n| flatfault-b  |     212 | 361.168 | 161.03   |  92.92 | 314.885  | 354.405 | 397.238 | 1121.32 |\n| curvefault-b |     212 | 450.822 | 182.416  | 176.84 | 396.375  | 430.89  | 464.228 | 1345.67 |\n| flatvel-b    |     212 | 578.099 | 247.847  |  26.37 | 577.227  | 649.205 | 705.645 | 1056.44 |\n| curvevel-b   |     212 | 645.475 | 174.971  | 123.56 | 623.84   | 670.83  | 729.697 | 1015.51 |\n\n---\n\n# [Unified Framework for Forward and Inverse Problems](https://github.com/KGML-lab/Generalized-Forward-Inverse-Framework-for-DL4SI)\n[Source](https://github.com/KGML-lab/Generalized-Forward-Inverse-Framework-for-DL4SI/tree/main/evaluate/Metrics_final)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2Feb0f8613ecf69a7319e3b6af4c85a8af%2FScreenshot%202025-04-27%20at%2012.50.50AM.png?generation=1745736307342753&alt=media)\n## FLATVEL-A \n\n| Model                                 | Source    | Target    |   MAE |\n|:--------------------------------------|:----------|:----------|------:|\n| UNetInverseModel 33M Latent32         | flatvel-a | flatvel-a |  6.63 |\n| UNetInverseModel 33M Latent64 No Skip | flatvel-a | flatvel-a |  7.45 |\n| UNetInverseModel 17M Latent32         | flatvel-a | flatvel-a |  7.49 |\n| UNetInverseModel 17M Latent32 No Skip | flatvel-a | flatvel-a |  8.29 |\n| UNetInverseModel 33M Latent32 No Skip | flatvel-a | flatvel-a |  8.58 |\n| Invertible XNet Adam                  | flatvel-a | flatvel-a |  8.66 |\n| UNetInverseModel 33M                  | flatvel-a | flatvel-a |  9.01 |\n| InversionNet                          | flatvel-a | flatvel-a |  9.77 |\n| UNetInverseModel 17M Latent16         | flatvel-a | flatvel-a | 11.05 |\n| UNetInverseModel 33M Latent16         | flatvel-a | flatvel-a | 11.09 |\n| UNetInverseModel 17M Latent8          | flatvel-a | flatvel-a | 11.44 |\n| UNetInverseModel 17M NoSkip           | flatvel-a | flatvel-a | 12.11 |\n| AutoLinear Inversion                  | flatvel-a | flatvel-a | 12.16 |\n| UNetInverseModel 17M Latent16 No Skip | flatvel-a | flatvel-a | 13.88 |\n| UNetInverseModel 17M Latent64         | flatvel-a | flatvel-a | 15.74 |\n\n---\n\n ## FLATFAULT-A \n\n| Model                                 | Source      | Target      |   MAE |\n|:--------------------------------------|:------------|:------------|------:|\n| UNetInverseModel 33M Latent64         | flatfault-a | flatfault-a | 12.88 |\n| UNetInverseModel 17M Latent64         | flatfault-a | flatfault-a | 14.52 |\n| UNetInverseModel 17M Latent32         | flatfault-a | flatfault-a | 15.41 |\n| UNetInverseModel 17M                  | flatfault-a | flatfault-a | 15.53 |\n| UNetInverseModel 17M NoSkip           | flatfault-a | flatfault-a | 15.85 |\n| UNetInverseModel 17M Latent32 No Skip | flatfault-a | flatfault-a | 16.11 |\n| UNetInverseModel 33M Latent64 No Skip | flatfault-a | flatfault-a | 17.04 |\n| UNetInverseModel 33M Latent32 No Skip | flatfault-a | flatfault-a | 17.65 |\n| UNetInverseModel 17M Latent64 No Skip | flatfault-a | flatfault-a | 19.19 |\n| UNetInverseModel 17M Latent16 No Skip | flatfault-a | flatfault-a | 20.54 |\n| UNetInverseModel 17M Latent8          | flatfault-a | flatfault-a | 20.84 |\n| UNetInverseModel 33M Latent16         | flatfault-a | flatfault-a | 21.7  |\n| UNetInverseModel 17M Latent16         | flatfault-a | flatfault-a | 23.18 |\n| UNetInverseModel 33M NoSkip           | flatfault-a | flatfault-a | 23.78 |\n| AutoLinear Inversion                  | flatfault-a | flatfault-a | 24.56 |\n\n---\n\n ## CURVEFAULT-A \n\n| Model                                 | Source       | Target       |   MAE |\n|:--------------------------------------|:-------------|:-------------|------:|\n| UNetInverseModel 33M Latent64         | curvefault-a | curvefault-a | 22.44 |\n| UNetInverseModel 33M NoSkip           | curvefault-a | curvefault-a | 23.4  |\n| UNetInverseModel 33M Latent64 No Skip | curvefault-a | curvefault-a | 26.42 |\n| UNetInverseModel 17M Latent32         | curvefault-a | curvefault-a | 26.98 |\n| UNetInverseModel 33M                  | curvefault-a | curvefault-a | 27.05 |\n| UNetInverseModel 33M Latent32         | curvefault-a | curvefault-a | 27.05 |\n| UNetInverseModel 17M                  | curvefault-a | curvefault-a | 28    |\n| UNetInverseModel 17M Latent32 No Skip | curvefault-a | curvefault-a | 30.12 |\n| UNetInverseModel 17M Latent16         | curvefault-a | curvefault-a | 32.53 |\n| UNetInverseModel 33M Latent16         | curvefault-a | curvefault-a | 33.53 |\n| UNetInverseModel 17M Latent16 No Skip | curvefault-a | curvefault-a | 34.29 |\n| UNetInverseModel 17M Latent64         | curvefault-a | curvefault-a | 35.3  |\n| UNetInverseModel 17M NoSkip           | curvefault-a | curvefault-a | 37.8  |\n| UNetInverseModel 33M Latent32 No Skip | curvefault-a | curvefault-a | 38.49 |\n| Velocity GAN                          | curvefault-a | curvefault-a | 38.8  |\n\n---\n\n ## FLATVEL-B \n\n| Model                         | Source    | Target    |   MAE |\n|:------------------------------|:----------|:----------|------:|\n| UNetInverseModel 33M Latent64 | flatvel-b | flatvel-b | 26.37 |\n| UNetInverseModel 33M Latent32 | flatvel-b | flatvel-b | 26.72 |\n| UNetInverseModel 33M NoSkip   | flatvel-b | flatvel-b | 27.1  |\n| UNetInverseModel 17M          | flatvel-b | flatvel-b | 27.95 |\n| UNetInverseModel 33M          | flatvel-b | flatvel-b | 28.12 |\n| UNetInverseModel 17M NoSkip   | flatvel-b | flatvel-b | 29    |\n| UNetInverseModel 17M Latent64 | flatvel-b | flatvel-b | 29.38 |\n| UNetInverseModel 17M Latent32 | flatvel-b | flatvel-b | 31.42 |\n| Invertible XNet Adam          | flatvel-b | flatvel-b | 33.16 |\n| InversionNet                  | flatvel-b | flatvel-b | 33.76 |\n| Invertible XNet               | flatvel-b | flatvel-b | 34.95 |\n| UNetInverseModel 33M Latent16 | flatvel-b | flatvel-b | 38.65 |\n| UNetInverseModel 17M Latent16 | flatvel-b | flatvel-b | 40.13 |\n| Invertible XNet cycle warmup  | flatvel-b | flatvel-b | 42.97 |\n| UNetInverseModel 33M Latent8  | flatvel-b | flatvel-b | 43.39 |\n\n---\n\n ## CURVEVEL-A \n\n| Model                                 | Source     | Target     |   MAE |\n|:--------------------------------------|:-----------|:-----------|------:|\n| UNetInverseModel 33M Latent64         | curvevel-a | curvevel-a | 46    |\n| UNetInverseModel 33M                  | curvevel-a | curvevel-a | 48.06 |\n| UNetInverseModel 33M NoSkip           | curvevel-a | curvevel-a | 49.63 |\n| UNetInverseModel 17M                  | curvevel-a | curvevel-a | 53.19 |\n| UNetInverseModel 33M Latent64 No Skip | curvevel-a | curvevel-a | 54.42 |\n| UNetInverseModel 17M Latent64         | curvevel-a | curvevel-a | 55.61 |\n| UNetInverseModel 33M Latent32         | curvevel-a | curvevel-a | 56.57 |\n| Invertible XNet                       | curvevel-a | curvevel-a | 57.87 |\n| UNetInverseModel 17M Latent32         | curvevel-a | curvevel-a | 59.94 |\n| UNetInverseModel 17M Latent64 No Skip | curvevel-a | curvevel-a | 60.99 |\n| Invertible XNet Adam                  | curvevel-a | curvevel-a | 62.15 |\n| UNetInverseModel 17M NoSkip           | curvevel-a | curvevel-a | 62.92 |\n| IUnetInverseModel                     | curvevel-a | curvevel-a | 62.92 |\n| Invertible XNet cycle warmup          | curvevel-a | curvevel-a | 63.04 |\n| UNetInverseModel 17M Latent32 No Skip | curvevel-a | curvevel-a | 63.47 |\n\n---\n\n ## STYLE-A \n\n| Model                        | Source   | Target   |    MAE |\n|:-----------------------------|:---------|:---------|-------:|\n| UNetInverseModel 33M         | style-a  | style-a  |  67    |\n| Invertible XNet              | style-a  | style-a  |  67.09 |\n| IUnetInverseModel            | style-a  | style-a  |  74.07 |\n| Invertible XNet cycle warmup | style-a  | style-a  |  81.71 |\n| Invertible XNet Adam         | style-a  | style-a  |  84.35 |\n| Velocity GAN                 | style-a  | style-a  |  91.77 |\n| InversionNet                 | style-a  | style-a  |  93.96 |\n| UNetInverseModel 17M         | style-a  | style-a  |  97.98 |\n| InversionNet                 | style-a  | style-a  | 103.36 |\n| AutoLinear Inversion         | style-a  | style-a  | 107.85 |\n| UNetInverseModel 33M         | style-b  | style-a  | 186.53 |\n| InversionNet                 | style-b  | style-a  | 187.93 |\n| UNetInverseModel 17M         | style-b  | style-a  | 192.72 |\n| Invertible XNet Adam         | style-b  | style-a  | 194.95 |\n| Invertible XNet              | style-b  | style-a  | 196.8  |\n\n---\n\n ## STYLE-B \n\n| Model                        | Source   | Target   |    MAE |\n|:-----------------------------|:---------|:---------|-------:|\n| UNetInverseModel 33M         | style-b  | style-b  |  71.41 |\n| UNetInverseModel 17M         | style-b  | style-b  |  83.62 |\n| Invertible XNet Adam         | style-b  | style-b  |  84.81 |\n| AutoLinear Inversion         | style-b  | style-b  |  95.63 |\n| InversionNet                 | style-b  | style-b  | 103.38 |\n| Velocity GAN                 | style-b  | style-b  | 104.52 |\n| InversionNet                 | style-b  | style-b  | 106.31 |\n| UNetInverseModel 33M         | style-a  | style-b  | 110.17 |\n| Invertible XNet              | style-a  | style-b  | 112.08 |\n| Invertible XNet Adam         | style-a  | style-b  | 120.73 |\n| IUnetInverseModel            | style-a  | style-b  | 120.96 |\n| Invertible XNet              | style-b  | style-b  | 121.43 |\n| Invertible XNet cycle warmup | style-a  | style-b  | 122.41 |\n| Invertible XNet cycle warmup | style-b  | style-b  | 129.17 |\n| InversionNet                 | style-a  | style-b  | 132.56 |\n\n---\n\n ## FLATFAULT-B \n\n| Model                         | Source      | Target      |    MAE |\n|:------------------------------|:------------|:------------|-------:|\n| UNetInverseModel 33M          | flatfault-b | flatfault-b |  92.92 |\n| Invertible XNet               | flatfault-b | flatfault-b |  94.12 |\n| IUnetInverseModel             | flatfault-b | flatfault-b |  95.2  |\n| Invertible XNet cycle warmup  | flatfault-b | flatfault-b |  95.24 |\n| UNetInverseModel 33M Latent64 | flatfault-b | flatfault-b |  95.28 |\n| UNetInverseModel 33M NoSkip   | flatfault-b | flatfault-b |  99.48 |\n| UNetInverseModel 33M Latent32 | flatfault-b | flatfault-b | 113.15 |\n| UNetInverseModel 17M Latent64 | flatfault-b | flatfault-b | 117.12 |\n| UNetInverseModel 17M          | flatfault-b | flatfault-b | 120.08 |\n| UNetInverseModel 17M NoSkip   | flatfault-b | flatfault-b | 122.25 |\n| UNetInverseModel 17M Latent32 | flatfault-b | flatfault-b | 125.66 |\n| Invertible XNet Adam          | flatfault-b | flatfault-b | 130.89 |\n| UNetInverseModel 33M Latent16 | flatfault-b | flatfault-b | 138.09 |\n| Velocity GAN                  | flatfault-b | flatfault-b | 138.89 |\n| InversionNet                  | flatfault-b | flatfault-b | 142.66 |\n\n---\n\n ## CURVEVEL-B \n\n| Model                         | Source     | Target     |    MAE |\n|:------------------------------|:-----------|:-----------|-------:|\n| UNetInverseModel 33M          | curvevel-b | curvevel-b | 123.56 |\n| UNetInverseModel 33M Latent64 | curvevel-b | curvevel-b | 126.07 |\n| IUnetInverseModel             | curvevel-b | curvevel-b | 135.35 |\n| UNetInverseModel 33M NoSkip   | curvevel-b | curvevel-b | 139.02 |\n| Invertible XNet               | curvevel-b | curvevel-b | 139.82 |\n| UNetInverseModel 33M Latent32 | curvevel-b | curvevel-b | 144.56 |\n| Invertible XNet cycle warmup  | curvevel-b | curvevel-b | 147.12 |\n| UNetInverseModel 17M Latent64 | curvevel-b | curvevel-b | 148.4  |\n| UNetInverseModel 17M          | curvevel-b | curvevel-b | 154.09 |\n| Invertible XNet Adam          | curvevel-b | curvevel-b | 156.69 |\n| UNetInverseModel 17M Latent32 | curvevel-b | curvevel-b | 157.16 |\n| UNetInverseModel 33M Latent16 | curvevel-b | curvevel-b | 166.42 |\n| UNetInverseModel 17M NoSkip   | curvevel-b | curvevel-b | 171.53 |\n| UNetInverseModel 17M Latent16 | curvevel-b | curvevel-b | 174.22 |\n| InversionNet                  | curvevel-b | curvevel-b | 182.81 |\n\n---\n\n ## CURVEFAULT-B \n\n| Model                         | Source       | Target       |    MAE |\n|:------------------------------|:-------------|:-------------|-------:|\n| IUnetInverseModel             | curvefault-b | curvefault-b | 176.84 |\n| Invertible XNet cycle warmup  | curvefault-b | curvefault-b | 177.19 |\n| Invertible XNet               | curvefault-b | curvefault-b | 181.03 |\n| UNetInverseModel 33M          | curvefault-b | curvefault-b | 185.23 |\n| UNetInverseModel 33M Latent64 | curvefault-b | curvefault-b | 188.04 |\n| UNetInverseModel 33M NoSkip   | curvefault-b | curvefault-b | 202.36 |\n| UNetInverseModel 33M Latent32 | curvefault-b | curvefault-b | 205.03 |\n| UNetInverseModel 17M Latent64 | curvefault-b | curvefault-b | 209.92 |\n| UNetInverseModel 17M Latent32 | curvefault-b | curvefault-b | 218.11 |\n| UNetInverseModel 17M          | curvefault-b | curvefault-b | 223.76 |\n| UNetInverseModel 17M NoSkip   | curvefault-b | curvefault-b | 223.81 |\n| Invertible XNet Adam          | curvefault-b | curvefault-b | 223.86 |\n| InversionNet                  | curvefault-b | curvefault-b | 228.18 |\n| UNetInverseModel 33M Latent16 | curvefault-b | curvefault-b | 229.39 |\n| Velocity GAN                  | curvefault-b | curvefault-b | 235.7  |",
      "votes": 17
    },
    {
      "id": 3188320,
      "postDate": "2025-04-27T12:16:49.037Z",
      "content": "<p>For anyone wondering, my CV scores for 43.6 LB</p>\n<table>\n<thead>\n<tr>\n<th>Method</th>\n<th>Score</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>CurveVel_A</td>\n<td>24.772050857543945</td>\n</tr>\n<tr>\n<td>CurveVel_B</td>\n<td>70.62040710449219</td>\n</tr>\n<tr>\n<td>FlatVel_A</td>\n<td>1.8052632808685303</td>\n</tr>\n<tr>\n<td>FlatVel_B</td>\n<td>10.19217300415039</td>\n</tr>\n<tr>\n<td>CurveFault_A</td>\n<td>7.599456787109375</td>\n</tr>\n<tr>\n<td>CurveFault_B</td>\n<td>110.47376251220703</td>\n</tr>\n<tr>\n<td>FlatFault_A</td>\n<td>3.7741732597351074</td>\n</tr>\n<tr>\n<td>FlatFault_B</td>\n<td>38.12479019165039</td>\n</tr>\n<tr>\n<td>Style_A</td>\n<td>43.12234878540039</td>\n</tr>\n<tr>\n<td>Style_B</td>\n<td>54.987770080566406</td>\n</tr>\n<tr>\n<td>Weighted Average</td>\n<td>39.22</td>\n</tr>\n</tbody>\n</table>",
      "rawMarkdown": "For anyone wondering, my CV scores for 43.6 LB\n\n\n| Method         | Score              |\n| -------------- | ------------------ |\n| CurveVel_A     | 24.772050857543945 |\n| CurveVel_B     | 70.62040710449219  |\n| FlatVel_A      | 1.8052632808685303 |\n| FlatVel_B      | 10.19217300415039  |\n| CurveFault_A   | 7.599456787109375  |\n| CurveFault_B   | 110.47376251220703 |\n| FlatFault_A    | 3.7741732597351074 |\n| FlatFault_B    | 38.12479019165039  |\n| Style_A        | 43.12234878540039  |\n| Style_B        | 54.987770080566406 |\n| Weighted Average        | 39.22 |",
      "votes": 13,
      "replies": [
        {
          "id": 3188325,
          "postDate": "2025-04-27T12:24:21.183Z",
          "content": "<p>Wow <a href=\"https://www.kaggle.com/harshitsheoran\" target=\"_blank\">@harshitsheoran</a> <strong>FlatVel_A -&gt; 1.8</strong> . Thanks for sharing. what is your CV strategy? </p>",
          "rawMarkdown": "Wow @harshitsheoran **FlatVel_A -> 1.8** . Thanks for sharing. what is your CV strategy? ",
          "votes": 1,
          "replies": [
            {
              "id": 3188336,
              "postDate": "2025-04-27T12:44:50.627Z",
              "content": "<p>Hi <a href=\"https://www.kaggle.com/seshurajup\" target=\"_blank\">@seshurajup</a>,</p>\n<p>CV Strategy is StratifiedGroupKFold, stratified on the methods to split them equally in train/val split of each fold<br>\nand to group on the base files so that every sample (500 of them) in something like OpenFWI/CurveVel_B/data/data1.npy either stay in training or validation</p>",
              "rawMarkdown": "Hi @seshurajup,\n\nCV Strategy is StratifiedGroupKFold, stratified on the methods to split them equally in train/val split of each fold\nand to group on the base files so that every sample (500 of them) in something like OpenFWI/CurveVel_B/data/data1.npy either stay in training or validation",
              "votes": 4
            }
          ]
        },
        {
          "id": 3188331,
          "postDate": "2025-04-27T12:35:32.440Z",
          "content": "<p>CurveVel_A so much larger than CurveFault_A does not make sense. I recommend you check again your validation, data or pipeline…(otherwise I suspect you accidentally swapped the scores)</p>",
          "rawMarkdown": "CurveVel_A so much larger than CurveFault_A does not make sense. I recommend you check again your validation, data or pipeline...(otherwise I suspect you accidentally swapped the scores)",
          "votes": 2,
          "replies": [
            {
              "id": 3188341,
              "postDate": "2025-04-27T12:52:57.133Z",
              "content": "<p>Hi <a href=\"https://www.kaggle.com/shlomoron\" target=\"_blank\">@shlomoron</a>,</p>\n<p>For this submission, CurveVel_A and CurveFault_B were not trained together in the same training, the hypermeters and model was the same, but in every training, even on my recent experiments, CurveFault_A is lower than CurveVel_A, is that not the case for you?</p>\n<p>I can be wrong but my guess was that because we have a lot more samples of CurveFault_A that the score is lower, even if I train only CurveFault_A vs CurveVel_A, CurveFault_A scores lower</p>\n<p>But I will check my validation loop manually today regardless to make sure</p>\n<p>(lower is better)</p>",
              "rawMarkdown": "Hi @shlomoron,\n\nFor this submission, CurveVel_A and CurveFault_B were not trained together in the same training, the hypermeters and model was the same, but in every training, even on my recent experiments, CurveFault_A is lower than CurveVel_A, is that not the case for you?\n\nI can be wrong but my guess was that because we have a lot more samples of CurveFault_A that the score is lower, even if I train only CurveFault_A vs CurveVel_A, CurveFault_A scores lower\n\nBut I will check my validation loop manually today regardless to make sure\n\n(lower is better)",
              "votes": 3
            },
            {
              "id": 3188353,
              "postDate": "2025-04-27T13:07:11.307Z",
              "content": "<p>If you trained different models it may make sense. Still strange.   <br>\nI use only one model. for my LB I have CurveVel_A 23.49518, CurveFault_A 60.<br>\nYour CurveFault_A  is insane.  </p>",
              "rawMarkdown": "If you trained different models it may make sense. Still strange.   \nI use only one model. for my LB I have CurveVel_A 23.49518, CurveFault_A 60.\nYour CurveFault_A  is insane.  ",
              "votes": 3
            },
            {
              "id": 3188385,
              "postDate": "2025-04-27T13:39:25.577Z",
              "content": "<p>That's interesting, it's probably the difference of our approaches</p>\n<p>Manually checked:</p>\n<ol>\n<li>Ground truth from original non-processed OpenFWI</li>\n<li>Ground truth used to calculate score/loss </li>\n<li>Model's predictions… <br>\nEverything check's out</li>\n</ol>\n<p>Although I have only submitted once, my LB is near CV so I doubt anything is wrong in the code that I would not have found :)</p>\n<p>Also wondering How 1st place do it :D</p>",
              "rawMarkdown": "That's interesting, it's probably the difference of our approaches\n\nManually checked:\n1. Ground truth from original non-processed OpenFWI\n2. Ground truth used to calculate score/loss \n3. Model's predictions... \nEverything check's out\n\nAlthough I have only submitted once, my LB is near CV so I doubt anything is wrong in the code that I would not have found :)\n\nAlso wondering How 1st place do it :D",
              "votes": 3
            },
            {
              "id": 3188402,
              "postDate": "2025-04-27T14:19:21.637Z",
              "content": "<p>It's really interesting; I look forward to seeing your solution at this point in ~2 months.<br>\nPlease remember what model you use currently so you can share haha (since it will probably change a lot until comp' end)</p>",
              "rawMarkdown": "It's really interesting; I look forward to seeing your solution at this point in ~2 months.\nPlease remember what model you use currently so you can share haha (since it will probably change a lot until comp' end)",
              "votes": 2
            },
            {
              "id": 3188421,
              "postDate": "2025-04-27T14:54:28.210Z",
              "content": "<p>I will make sure to recall :)</p>\n<p>43.6 is actually much more simple of a solution than you might think, One can argue about how much time it takes to train something like that, I would say that it's possible to train a model a little under 40 LB in less than 6 hours on a single 3090 (it doesn't finish converging in just 6 hours), having a lot of compute does help but in my experiments, the trainings are very stable and even if we use less than full data, or smaller models, the correlation is very good when scaling up</p>",
              "rawMarkdown": "I will make sure to recall :)\n\n43.6 is actually much more simple of a solution than you might think, One can argue about how much time it takes to train something like that, I would say that it's possible to train a model a little under 40 LB in less than 6 hours on a single 3090 (it doesn't finish converging in just 6 hours), having a lot of compute does help but in my experiments, the trainings are very stable and even if we use less than full data, or smaller models, the correlation is very good when scaling up",
              "votes": 5
            },
            {
              "id": 3188427,
              "postDate": "2025-04-27T15:07:10.823Z",
              "content": "<blockquote>\n  <p>For this submission, CurveVel_A and CurveFault_B were not trained together in the same training,</p>\n</blockquote>\n<p><a href=\"https://www.kaggle.com/harshitsheoran\" target=\"_blank\">@harshitsheoran</a> are you using classification of the test samples to choice the dataset's model?</p>",
              "rawMarkdown": ">For this submission, CurveVel_A and CurveFault_B were not trained together in the same training,\n\n@harshitsheoran are you using classification of the test samples to choice the dataset's model?",
              "votes": 1
            },
            {
              "id": 3188434,
              "postDate": "2025-04-27T15:22:57.297Z",
              "content": "<p><a href=\"https://www.kaggle.com/seshurajup\" target=\"_blank\">@seshurajup</a> Yes, I did make a classification model for the 43.6 submission to choose which model to use for which sample, but since then I have shifted to a single model all data</p>",
              "rawMarkdown": "@seshurajup Yes, I did make a classification model for the 43.6 submission to choose which model to use for which sample, but since then I have shifted to a single model all data",
              "votes": 6
            },
            {
              "id": 3188437,
              "postDate": "2025-04-27T15:30:56.120Z",
              "content": "<blockquote>\n  <p>43.6 is actually much more simple of a solution than you might think  </p>\n</blockquote>\n<p>It's always simple after you find the solution.  <br>\nHowever, finding the simple solution is very hard 😀  </p>",
              "rawMarkdown": ">43.6 is actually much more simple of a solution than you might think  \n\nIt's always simple after you find the solution.  \nHowever, finding the simple solution is very hard 😀  ",
              "votes": 4
            },
            {
              "id": 3192858,
              "postDate": "2025-05-03T13:04:00.390Z",
              "content": "<p>Out of curiosity, is your current LB score from a single model trained on a subset of openfwi, or does it already use 100% of the data and ensembling to further boost scores? Wondering how far one can go without resorting to these measures on top.</p>",
              "rawMarkdown": "Out of curiosity, is your current LB score from a single model trained on a subset of openfwi, or does it already use 100% of the data and ensembling to further boost scores? Wondering how far one can go without resorting to these measures on top."
            }
          ]
        }
      ]
    },
    {
      "id": 3188642,
      "postDate": "2025-04-28T01:51:57.363Z",
      "content": "<p>All training data is too large. I wonder if there is a method that can use less data to get a good score.</p>",
      "rawMarkdown": "All training data is too large. I wonder if there is a method that can use less data to get a good score.",
      "votes": 3,
      "replies": [
        {
          "id": 3192797,
          "postDate": "2025-05-03T11:34:07.163Z",
          "content": "<p>Active learning</p>",
          "rawMarkdown": "Active learning"
        }
      ]
    },
    {
      "id": 3188206,
      "postDate": "2025-04-27T08:08:19.943Z",
      "content": "<p>What does this mean? Is this some benchmarking you did yourself? Or data extracted from the paper? </p>\n<p>Just my personal opinion, but posts without any context, just dumped data, don't really inspire discussions.</p>",
      "rawMarkdown": "What does this mean? Is this some benchmarking you did yourself? Or data extracted from the paper? \n\nJust my personal opinion, but posts without any context, just dumped data, don't really inspire discussions.",
      "votes": 1,
      "replies": [
        {
          "id": 3188212,
          "postDate": "2025-04-27T08:18:14.370Z",
          "content": "<p>its purely based on paper metric analysis. <a href=\"https://www.kaggle.com/fpeccia\" target=\"_blank\">@fpeccia</a> -&gt; from <a href=\"https://www.kaggle.com/competitions/waveform-inversion/discussion/572399\" target=\"_blank\">OpenFWI models results correlate with LB</a> are tested on LB, results are correlate with dataset models. </p>",
          "rawMarkdown": "its purely based on paper metric analysis. @fpeccia -> from [OpenFWI models results correlate with LB](https://www.kaggle.com/competitions/waveform-inversion/discussion/572399) are tested on LB, results are correlate with dataset models. "
        },
        {
          "id": 3188213,
          "postDate": "2025-04-27T08:18:40.843Z",
          "rawMarkdown": "",
          "isDeleted": true
        }
      ]
    },
    {
      "id": 3195167,
      "postDate": "2025-05-06T17:50:12.677Z",
      "content": "<p>I'm wondering how you transpose the dedicated trained models for different styles of velocity map to the unlabelled test data ? </p>",
      "rawMarkdown": "I'm wondering how you transpose the dedicated trained models for different styles of velocity map to the unlabelled test data ? "
    },
    {
      "id": 3197819,
      "postDate": "2025-05-08T16:32:20.117Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 3188320,
      "author_name": "Harshit Sheoran",
      "author_url": "",
      "post_date": "2025-04-27T12:16:49.037000",
      "content": "<p>For anyone wondering, my CV scores for 43.6 LB</p>\n<table>\n<thead>\n<tr>\n<th>Method</th>\n<th>Score</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>CurveVel_A</td>\n<td>24.772050857543945</td>\n</tr>\n<tr>\n<td>CurveVel_B</td>\n<td>70.62040710449219</td>\n</tr>\n<tr>\n<td>FlatVel_A</td>\n<td>1.8052632808685303</td>\n</tr>\n<tr>\n<td>FlatVel_B</td>\n<td>10.19217300415039</td>\n</tr>\n<tr>\n<td>CurveFault_A</td>\n<td>7.599456787109375</td>\n</tr>\n<tr>\n<td>CurveFault_B</td>\n<td>110.47376251220703</td>\n</tr>\n<tr>\n<td>FlatFault_A</td>\n<td>3.7741732597351074</td>\n</tr>\n<tr>\n<td>FlatFault_B</td>\n<td>38.12479019165039</td>\n</tr>\n<tr>\n<td>Style_A</td>\n<td>43.12234878540039</td>\n</tr>\n<tr>\n<td>Style_B</td>\n<td>54.987770080566406</td>\n</tr>\n<tr>\n<td>Weighted Average</td>\n<td>39.22</td>\n</tr>\n</tbody>\n</table>",
      "votes": 13,
      "replies": [
        {
          "id": 3188325,
          "author_name": "SeshuRaju 🧘‍♂️",
          "author_url": "",
          "post_date": "2025-04-27T12:24:21.183000",
          "content": "<p>Wow <a href=\"https://www.kaggle.com/harshitsheoran\" target=\"_blank\">@harshitsheoran</a> <strong>FlatVel_A -&gt; 1.8</strong> . Thanks for sharing. what is your CV strategy? </p>",
          "votes": 1,
          "replies": [
            {
              "id": 3188336,
              "author_name": "Harshit Sheoran",
              "author_url": "",
              "post_date": "2025-04-27T12:44:50.627000",
              "content": "<p>Hi <a href=\"https://www.kaggle.com/seshurajup\" target=\"_blank\">@seshurajup</a>,</p>\n<p>CV Strategy is StratifiedGroupKFold, stratified on the methods to split them equally in train/val split of each fold<br>\nand to group on the base files so that every sample (500 of them) in something like OpenFWI/CurveVel_B/data/data1.npy either stay in training or validation</p>",
              "votes": 4,
              "replies": []
            }
          ]
        },
        {
          "id": 3188331,
          "author_name": "greySnow",
          "author_url": "",
          "post_date": "2025-04-27T12:35:32.440000",
          "content": "<p>CurveVel_A so much larger than CurveFault_A does not make sense. I recommend you check again your validation, data or pipeline…(otherwise I suspect you accidentally swapped the scores)</p>",
          "votes": 2,
          "replies": [
            {
              "id": 3188341,
              "author_name": "Harshit Sheoran",
              "author_url": "",
              "post_date": "2025-04-27T12:52:57.133000",
              "content": "<p>Hi <a href=\"https://www.kaggle.com/shlomoron\" target=\"_blank\">@shlomoron</a>,</p>\n<p>For this submission, CurveVel_A and CurveFault_B were not trained together in the same training, the hypermeters and model was the same, but in every training, even on my recent experiments, CurveFault_A is lower than CurveVel_A, is that not the case for you?</p>\n<p>I can be wrong but my guess was that because we have a lot more samples of CurveFault_A that the score is lower, even if I train only CurveFault_A vs CurveVel_A, CurveFault_A scores lower</p>\n<p>But I will check my validation loop manually today regardless to make sure</p>\n<p>(lower is better)</p>",
              "votes": 3,
              "replies": []
            },
            {
              "id": 3188353,
              "author_name": "greySnow",
              "author_url": "",
              "post_date": "2025-04-27T13:07:11.307000",
              "content": "<p>If you trained different models it may make sense. Still strange.   <br>\nI use only one model. for my LB I have CurveVel_A 23.49518, CurveFault_A 60.<br>\nYour CurveFault_A  is insane.  </p>",
              "votes": 3,
              "replies": []
            },
            {
              "id": 3188385,
              "author_name": "Harshit Sheoran",
              "author_url": "",
              "post_date": "2025-04-27T13:39:25.577000",
              "content": "<p>That's interesting, it's probably the difference of our approaches</p>\n<p>Manually checked:</p>\n<ol>\n<li>Ground truth from original non-processed OpenFWI</li>\n<li>Ground truth used to calculate score/loss </li>\n<li>Model's predictions… <br>\nEverything check's out</li>\n</ol>\n<p>Although I have only submitted once, my LB is near CV so I doubt anything is wrong in the code that I would not have found :)</p>\n<p>Also wondering How 1st place do it :D</p>",
              "votes": 3,
              "replies": []
            },
            {
              "id": 3188402,
              "author_name": "greySnow",
              "author_url": "",
              "post_date": "2025-04-27T14:19:21.637000",
              "content": "<p>It's really interesting; I look forward to seeing your solution at this point in ~2 months.<br>\nPlease remember what model you use currently so you can share haha (since it will probably change a lot until comp' end)</p>",
              "votes": 2,
              "replies": []
            },
            {
              "id": 3188421,
              "author_name": "Harshit Sheoran",
              "author_url": "",
              "post_date": "2025-04-27T14:54:28.210000",
              "content": "<p>I will make sure to recall :)</p>\n<p>43.6 is actually much more simple of a solution than you might think, One can argue about how much time it takes to train something like that, I would say that it's possible to train a model a little under 40 LB in less than 6 hours on a single 3090 (it doesn't finish converging in just 6 hours), having a lot of compute does help but in my experiments, the trainings are very stable and even if we use less than full data, or smaller models, the correlation is very good when scaling up</p>",
              "votes": 5,
              "replies": []
            },
            {
              "id": 3188427,
              "author_name": "SeshuRaju 🧘‍♂️",
              "author_url": "",
              "post_date": "2025-04-27T15:07:10.823000",
              "content": "<blockquote>\n  <p>For this submission, CurveVel_A and CurveFault_B were not trained together in the same training,</p>\n</blockquote>\n<p><a href=\"https://www.kaggle.com/harshitsheoran\" target=\"_blank\">@harshitsheoran</a> are you using classification of the test samples to choice the dataset's model?</p>",
              "votes": 1,
              "replies": []
            },
            {
              "id": 3188434,
              "author_name": "Harshit Sheoran",
              "author_url": "",
              "post_date": "2025-04-27T15:22:57.297000",
              "content": "<p><a href=\"https://www.kaggle.com/seshurajup\" target=\"_blank\">@seshurajup</a> Yes, I did make a classification model for the 43.6 submission to choose which model to use for which sample, but since then I have shifted to a single model all data</p>",
              "votes": 6,
              "replies": []
            },
            {
              "id": 3188437,
              "author_name": "greySnow",
              "author_url": "",
              "post_date": "2025-04-27T15:30:56.120000",
              "content": "<blockquote>\n  <p>43.6 is actually much more simple of a solution than you might think  </p>\n</blockquote>\n<p>It's always simple after you find the solution.  <br>\nHowever, finding the simple solution is very hard 😀  </p>",
              "votes": 4,
              "replies": []
            },
            {
              "id": 3192858,
              "author_name": "Sebastian Hoffmann",
              "author_url": "",
              "post_date": "2025-05-03T13:04:00.390000",
              "content": "<p>Out of curiosity, is your current LB score from a single model trained on a subset of openfwi, or does it already use 100% of the data and ensembling to further boost scores? Wondering how far one can go without resorting to these measures on top.</p>",
              "votes": 0,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 3188642,
      "author_name": "I2nfinit3y",
      "author_url": "",
      "post_date": "2025-04-28T01:51:57.363000",
      "content": "<p>All training data is too large. I wonder if there is a method that can use less data to get a good score.</p>",
      "votes": 3,
      "replies": [
        {
          "id": 3192797,
          "author_name": "Aleksey Trepetsky",
          "author_url": "",
          "post_date": "2025-05-03T11:34:07.163000",
          "content": "<p>Active learning</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 3188206,
      "author_name": "Federico Peccia",
      "author_url": "",
      "post_date": "2025-04-27T08:08:19.943000",
      "content": "<p>What does this mean? Is this some benchmarking you did yourself? Or data extracted from the paper? </p>\n<p>Just my personal opinion, but posts without any context, just dumped data, don't really inspire discussions.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 3188212,
          "author_name": "SeshuRaju 🧘‍♂️",
          "author_url": "",
          "post_date": "2025-04-27T08:18:14.370000",
          "content": "<p>its purely based on paper metric analysis. <a href=\"https://www.kaggle.com/fpeccia\" target=\"_blank\">@fpeccia</a> -&gt; from <a href=\"https://www.kaggle.com/competitions/waveform-inversion/discussion/572399\" target=\"_blank\">OpenFWI models results correlate with LB</a> are tested on LB, results are correlate with dataset models. </p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 3188213,
          "author_name": "",
          "author_url": "",
          "post_date": "2025-04-27T08:18:40.843000",
          "content": "",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 3195167,
      "author_name": "Kanadry",
      "author_url": "",
      "post_date": "2025-05-06T17:50:12.677000",
      "content": "<p>I'm wondering how you transpose the dedicated trained models for different styles of velocity map to the unlabelled test data ? </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 3197819,
      "author_name": "",
      "author_url": "",
      "post_date": "2025-05-08T16:32:20.117000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3188167": "# Average of all best individual models -> **MAE - 64.60 for DL4SI models**\n>| Model                         | Source       | Target       |    MAE |\n|:------------------------------|:-------------|:-------------|-------:|\n| UNetInverseModel 33M Latent32 | flatvel-a    | flatvel-a    |   6.63 |\n| UNetInverseModel 33M Latent64 | flatfault-a  | flatfault-a  |  12.88 |\n| UNetInverseModel 33M Latent64 | curvefault-a | curvefault-a |  22.44 |\n| UNetInverseModel 33M Latent64 | flatvel-b    | flatvel-b    |  26.37 |\n| UNetInverseModel 33M Latent64 | curvevel-a   | curvevel-a   |  46.00    |\n| UNetInverseModel 33M          | style-a      | style-a      |  67.00    |\n| UNetInverseModel 33M          | style-b      | style-b      |  71.41 |\n| UNetInverseModel 33M          | flatfault-b  | flatfault-b  |  92.92 |\n| UNetInverseModel 33M          | curvevel-b   | curvevel-b   | 123.56 |\n| IUnetInverseModel             | curvefault-b | curvefault-b | 176.84 |\n\n---\n\n| Target       |   count |    mean |      std |    min |      25% |     50% |     75% |     max |\n|:-------------|--------:|--------:|---------:|-------:|---------:|--------:|--------:|--------:|\n| flatvel-a    |     212 | 115.487 |  94.1268 |   6.63 |  51.4    |  95.255 | 144.572 |  502.67 |\n| flatfault-a  |     212 | 151.515 | 112.252  |  12.88 |  75.7225 | 116.315 | 191.968 |  635.85 |\n| curvefault-a |     212 | 242.581 | 180.301  |  22.44 | 135.422  | 176.475 | 357.52  | 1101.14 |\n| curvevel-a   |     212 | 244.101 | 140.08   |  46    | 162.987  | 223.74  | 297.4   |  889.18 |\n| style-b      |     212 | 269.586 | 138.38   |  71.41 | 213.805  | 239.845 | 284.812 | 1012.49 |\n| style-a      |     212 | 347.427 | 134.048  |  67    | 281.2    | 330.29  | 376.275 |  894.53 |\n| flatfault-b  |     212 | 361.168 | 161.03   |  92.92 | 314.885  | 354.405 | 397.238 | 1121.32 |\n| curvefault-b |     212 | 450.822 | 182.416  | 176.84 | 396.375  | 430.89  | 464.228 | 1345.67 |\n| flatvel-b    |     212 | 578.099 | 247.847  |  26.37 | 577.227  | 649.205 | 705.645 | 1056.44 |\n| curvevel-b   |     212 | 645.475 | 174.971  | 123.56 | 623.84   | 670.83  | 729.697 | 1015.51 |\n\n---\n\n# [Unified Framework for Forward and Inverse Problems](https://github.com/KGML-lab/Generalized-Forward-Inverse-Framework-for-DL4SI)\n[Source](https://github.com/KGML-lab/Generalized-Forward-Inverse-Framework-for-DL4SI/tree/main/evaluate/Metrics_final)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2Feb0f8613ecf69a7319e3b6af4c85a8af%2FScreenshot%202025-04-27%20at%2012.50.50AM.png?generation=1745736307342753&alt=media)\n## FLATVEL-A \n\n| Model                                 | Source    | Target    |   MAE |\n|:--------------------------------------|:----------|:----------|------:|\n| UNetInverseModel 33M Latent32         | flatvel-a | flatvel-a |  6.63 |\n| UNetInverseModel 33M Latent64 No Skip | flatvel-a | flatvel-a |  7.45 |\n| UNetInverseModel 17M Latent32         | flatvel-a | flatvel-a |  7.49 |\n| UNetInverseModel 17M Latent32 No Skip | flatvel-a | flatvel-a |  8.29 |\n| UNetInverseModel 33M Latent32 No Skip | flatvel-a | flatvel-a |  8.58 |\n| Invertible XNet Adam                  | flatvel-a | flatvel-a |  8.66 |\n| UNetInverseModel 33M                  | flatvel-a | flatvel-a |  9.01 |\n| InversionNet                          | flatvel-a | flatvel-a |  9.77 |\n| UNetInverseModel 17M Latent16         | flatvel-a | flatvel-a | 11.05 |\n| UNetInverseModel 33M Latent16         | flatvel-a | flatvel-a | 11.09 |\n| UNetInverseModel 17M Latent8          | flatvel-a | flatvel-a | 11.44 |\n| UNetInverseModel 17M NoSkip           | flatvel-a | flatvel-a | 12.11 |\n| AutoLinear Inversion                  | flatvel-a | flatvel-a | 12.16 |\n| UNetInverseModel 17M Latent16 No Skip | flatvel-a | flatvel-a | 13.88 |\n| UNetInverseModel 17M Latent64         | flatvel-a | flatvel-a | 15.74 |\n\n---\n\n ## FLATFAULT-A \n\n| Model                                 | Source      | Target      |   MAE |\n|:--------------------------------------|:------------|:------------|------:|\n| UNetInverseModel 33M Latent64         | flatfault-a | flatfault-a | 12.88 |\n| UNetInverseModel 17M Latent64         | flatfault-a | flatfault-a | 14.52 |\n| UNetInverseModel 17M Latent32         | flatfault-a | flatfault-a | 15.41 |\n| UNetInverseModel 17M                  | flatfault-a | flatfault-a | 15.53 |\n| UNetInverseModel 17M NoSkip           | flatfault-a | flatfault-a | 15.85 |\n| UNetInverseModel 17M Latent32 No Skip | flatfault-a | flatfault-a | 16.11 |\n| UNetInverseModel 33M Latent64 No Skip | flatfault-a | flatfault-a | 17.04 |\n| UNetInverseModel 33M Latent32 No Skip | flatfault-a | flatfault-a | 17.65 |\n| UNetInverseModel 17M Latent64 No Skip | flatfault-a | flatfault-a | 19.19 |\n| UNetInverseModel 17M Latent16 No Skip | flatfault-a | flatfault-a | 20.54 |\n| UNetInverseModel 17M Latent8          | flatfault-a | flatfault-a | 20.84 |\n| UNetInverseModel 33M Latent16         | flatfault-a | flatfault-a | 21.7  |\n| UNetInverseModel 17M Latent16         | flatfault-a | flatfault-a | 23.18 |\n| UNetInverseModel 33M NoSkip           | flatfault-a | flatfault-a | 23.78 |\n| AutoLinear Inversion                  | flatfault-a | flatfault-a | 24.56 |\n\n---\n\n ## CURVEFAULT-A \n\n| Model                                 | Source       | Target       |   MAE |\n|:--------------------------------------|:-------------|:-------------|------:|\n| UNetInverseModel 33M Latent64         | curvefault-a | curvefault-a | 22.44 |\n| UNetInverseModel 33M NoSkip           | curvefault-a | curvefault-a | 23.4  |\n| UNetInverseModel 33M Latent64 No Skip | curvefault-a | curvefault-a | 26.42 |\n| UNetInverseModel 17M Latent32         | curvefault-a | curvefault-a | 26.98 |\n| UNetInverseModel 33M                  | curvefault-a | curvefault-a | 27.05 |\n| UNetInverseModel 33M Latent32         | curvefault-a | curvefault-a | 27.05 |\n| UNetInverseModel 17M                  | curvefault-a | curvefault-a | 28    |\n| UNetInverseModel 17M Latent32 No Skip | curvefault-a | curvefault-a | 30.12 |\n| UNetInverseModel 17M Latent16         | curvefault-a | curvefault-a | 32.53 |\n| UNetInverseModel 33M Latent16         | curvefault-a | curvefault-a | 33.53 |\n| UNetInverseModel 17M Latent16 No Skip | curvefault-a | curvefault-a | 34.29 |\n| UNetInverseModel 17M Latent64         | curvefault-a | curvefault-a | 35.3  |\n| UNetInverseModel 17M NoSkip           | curvefault-a | curvefault-a | 37.8  |\n| UNetInverseModel 33M Latent32 No Skip | curvefault-a | curvefault-a | 38.49 |\n| Velocity GAN                          | curvefault-a | curvefault-a | 38.8  |\n\n---\n\n ## FLATVEL-B \n\n| Model                         | Source    | Target    |   MAE |\n|:------------------------------|:----------|:----------|------:|\n| UNetInverseModel 33M Latent64 | flatvel-b | flatvel-b | 26.37 |\n| UNetInverseModel 33M Latent32 | flatvel-b | flatvel-b | 26.72 |\n| UNetInverseModel 33M NoSkip   | flatvel-b | flatvel-b | 27.1  |\n| UNetInverseModel 17M          | flatvel-b | flatvel-b | 27.95 |\n| UNetInverseModel 33M          | flatvel-b | flatvel-b | 28.12 |\n| UNetInverseModel 17M NoSkip   | flatvel-b | flatvel-b | 29    |\n| UNetInverseModel 17M Latent64 | flatvel-b | flatvel-b | 29.38 |\n| UNetInverseModel 17M Latent32 | flatvel-b | flatvel-b | 31.42 |\n| Invertible XNet Adam          | flatvel-b | flatvel-b | 33.16 |\n| InversionNet                  | flatvel-b | flatvel-b | 33.76 |\n| Invertible XNet               | flatvel-b | flatvel-b | 34.95 |\n| UNetInverseModel 33M Latent16 | flatvel-b | flatvel-b | 38.65 |\n| UNetInverseModel 17M Latent16 | flatvel-b | flatvel-b | 40.13 |\n| Invertible XNet cycle warmup  | flatvel-b | flatvel-b | 42.97 |\n| UNetInverseModel 33M Latent8  | flatvel-b | flatvel-b | 43.39 |\n\n---\n\n ## CURVEVEL-A \n\n| Model                                 | Source     | Target     |   MAE |\n|:--------------------------------------|:-----------|:-----------|------:|\n| UNetInverseModel 33M Latent64         | curvevel-a | curvevel-a | 46    |\n| UNetInverseModel 33M                  | curvevel-a | curvevel-a | 48.06 |\n| UNetInverseModel 33M NoSkip           | curvevel-a | curvevel-a | 49.63 |\n| UNetInverseModel 17M                  | curvevel-a | curvevel-a | 53.19 |\n| UNetInverseModel 33M Latent64 No Skip | curvevel-a | curvevel-a | 54.42 |\n| UNetInverseModel 17M Latent64         | curvevel-a | curvevel-a | 55.61 |\n| UNetInverseModel 33M Latent32         | curvevel-a | curvevel-a | 56.57 |\n| Invertible XNet                       | curvevel-a | curvevel-a | 57.87 |\n| UNetInverseModel 17M Latent32         | curvevel-a | curvevel-a | 59.94 |\n| UNetInverseModel 17M Latent64 No Skip | curvevel-a | curvevel-a | 60.99 |\n| Invertible XNet Adam                  | curvevel-a | curvevel-a | 62.15 |\n| UNetInverseModel 17M NoSkip           | curvevel-a | curvevel-a | 62.92 |\n| IUnetInverseModel                     | curvevel-a | curvevel-a | 62.92 |\n| Invertible XNet cycle warmup          | curvevel-a | curvevel-a | 63.04 |\n| UNetInverseModel 17M Latent32 No Skip | curvevel-a | curvevel-a | 63.47 |\n\n---\n\n ## STYLE-A \n\n| Model                        | Source   | Target   |    MAE |\n|:-----------------------------|:---------|:---------|-------:|\n| UNetInverseModel 33M         | style-a  | style-a  |  67    |\n| Invertible XNet              | style-a  | style-a  |  67.09 |\n| IUnetInverseModel            | style-a  | style-a  |  74.07 |\n| Invertible XNet cycle warmup | style-a  | style-a  |  81.71 |\n| Invertible XNet Adam         | style-a  | style-a  |  84.35 |\n| Velocity GAN                 | style-a  | style-a  |  91.77 |\n| InversionNet                 | style-a  | style-a  |  93.96 |\n| UNetInverseModel 17M         | style-a  | style-a  |  97.98 |\n| InversionNet                 | style-a  | style-a  | 103.36 |\n| AutoLinear Inversion         | style-a  | style-a  | 107.85 |\n| UNetInverseModel 33M         | style-b  | style-a  | 186.53 |\n| InversionNet                 | style-b  | style-a  | 187.93 |\n| UNetInverseModel 17M         | style-b  | style-a  | 192.72 |\n| Invertible XNet Adam         | style-b  | style-a  | 194.95 |\n| Invertible XNet              | style-b  | style-a  | 196.8  |\n\n---\n\n ## STYLE-B \n\n| Model                        | Source   | Target   |    MAE |\n|:-----------------------------|:---------|:---------|-------:|\n| UNetInverseModel 33M         | style-b  | style-b  |  71.41 |\n| UNetInverseModel 17M         | style-b  | style-b  |  83.62 |\n| Invertible XNet Adam         | style-b  | style-b  |  84.81 |\n| AutoLinear Inversion         | style-b  | style-b  |  95.63 |\n| InversionNet                 | style-b  | style-b  | 103.38 |\n| Velocity GAN                 | style-b  | style-b  | 104.52 |\n| InversionNet                 | style-b  | style-b  | 106.31 |\n| UNetInverseModel 33M         | style-a  | style-b  | 110.17 |\n| Invertible XNet              | style-a  | style-b  | 112.08 |\n| Invertible XNet Adam         | style-a  | style-b  | 120.73 |\n| IUnetInverseModel            | style-a  | style-b  | 120.96 |\n| Invertible XNet              | style-b  | style-b  | 121.43 |\n| Invertible XNet cycle warmup | style-a  | style-b  | 122.41 |\n| Invertible XNet cycle warmup | style-b  | style-b  | 129.17 |\n| InversionNet                 | style-a  | style-b  | 132.56 |\n\n---\n\n ## FLATFAULT-B \n\n| Model                         | Source      | Target      |    MAE |\n|:------------------------------|:------------|:------------|-------:|\n| UNetInverseModel 33M          | flatfault-b | flatfault-b |  92.92 |\n| Invertible XNet               | flatfault-b | flatfault-b |  94.12 |\n| IUnetInverseModel             | flatfault-b | flatfault-b |  95.2  |\n| Invertible XNet cycle warmup  | flatfault-b | flatfault-b |  95.24 |\n| UNetInverseModel 33M Latent64 | flatfault-b | flatfault-b |  95.28 |\n| UNetInverseModel 33M NoSkip   | flatfault-b | flatfault-b |  99.48 |\n| UNetInverseModel 33M Latent32 | flatfault-b | flatfault-b | 113.15 |\n| UNetInverseModel 17M Latent64 | flatfault-b | flatfault-b | 117.12 |\n| UNetInverseModel 17M          | flatfault-b | flatfault-b | 120.08 |\n| UNetInverseModel 17M NoSkip   | flatfault-b | flatfault-b | 122.25 |\n| UNetInverseModel 17M Latent32 | flatfault-b | flatfault-b | 125.66 |\n| Invertible XNet Adam          | flatfault-b | flatfault-b | 130.89 |\n| UNetInverseModel 33M Latent16 | flatfault-b | flatfault-b | 138.09 |\n| Velocity GAN                  | flatfault-b | flatfault-b | 138.89 |\n| InversionNet                  | flatfault-b | flatfault-b | 142.66 |\n\n---\n\n ## CURVEVEL-B \n\n| Model                         | Source     | Target     |    MAE |\n|:------------------------------|:-----------|:-----------|-------:|\n| UNetInverseModel 33M          | curvevel-b | curvevel-b | 123.56 |\n| UNetInverseModel 33M Latent64 | curvevel-b | curvevel-b | 126.07 |\n| IUnetInverseModel             | curvevel-b | curvevel-b | 135.35 |\n| UNetInverseModel 33M NoSkip   | curvevel-b | curvevel-b | 139.02 |\n| Invertible XNet               | curvevel-b | curvevel-b | 139.82 |\n| UNetInverseModel 33M Latent32 | curvevel-b | curvevel-b | 144.56 |\n| Invertible XNet cycle warmup  | curvevel-b | curvevel-b | 147.12 |\n| UNetInverseModel 17M Latent64 | curvevel-b | curvevel-b | 148.4  |\n| UNetInverseModel 17M          | curvevel-b | curvevel-b | 154.09 |\n| Invertible XNet Adam          | curvevel-b | curvevel-b | 156.69 |\n| UNetInverseModel 17M Latent32 | curvevel-b | curvevel-b | 157.16 |\n| UNetInverseModel 33M Latent16 | curvevel-b | curvevel-b | 166.42 |\n| UNetInverseModel 17M NoSkip   | curvevel-b | curvevel-b | 171.53 |\n| UNetInverseModel 17M Latent16 | curvevel-b | curvevel-b | 174.22 |\n| InversionNet                  | curvevel-b | curvevel-b | 182.81 |\n\n---\n\n ## CURVEFAULT-B \n\n| Model                         | Source       | Target       |    MAE |\n|:------------------------------|:-------------|:-------------|-------:|\n| IUnetInverseModel             | curvefault-b | curvefault-b | 176.84 |\n| Invertible XNet cycle warmup  | curvefault-b | curvefault-b | 177.19 |\n| Invertible XNet               | curvefault-b | curvefault-b | 181.03 |\n| UNetInverseModel 33M          | curvefault-b | curvefault-b | 185.23 |\n| UNetInverseModel 33M Latent64 | curvefault-b | curvefault-b | 188.04 |\n| UNetInverseModel 33M NoSkip   | curvefault-b | curvefault-b | 202.36 |\n| UNetInverseModel 33M Latent32 | curvefault-b | curvefault-b | 205.03 |\n| UNetInverseModel 17M Latent64 | curvefault-b | curvefault-b | 209.92 |\n| UNetInverseModel 17M Latent32 | curvefault-b | curvefault-b | 218.11 |\n| UNetInverseModel 17M          | curvefault-b | curvefault-b | 223.76 |\n| UNetInverseModel 17M NoSkip   | curvefault-b | curvefault-b | 223.81 |\n| Invertible XNet Adam          | curvefault-b | curvefault-b | 223.86 |\n| InversionNet                  | curvefault-b | curvefault-b | 228.18 |\n| UNetInverseModel 33M Latent16 | curvefault-b | curvefault-b | 229.39 |\n| Velocity GAN                  | curvefault-b | curvefault-b | 235.7  |",
    "3188320": "For anyone wondering, my CV scores for 43.6 LB\n\n\n| Method         | Score              |\n| -------------- | ------------------ |\n| CurveVel_A     | 24.772050857543945 |\n| CurveVel_B     | 70.62040710449219  |\n| FlatVel_A      | 1.8052632808685303 |\n| FlatVel_B      | 10.19217300415039  |\n| CurveFault_A   | 7.599456787109375  |\n| CurveFault_B   | 110.47376251220703 |\n| FlatFault_A    | 3.7741732597351074 |\n| FlatFault_B    | 38.12479019165039  |\n| Style_A        | 43.12234878540039  |\n| Style_B        | 54.987770080566406 |\n| Weighted Average        | 39.22 |",
    "3188642": "All training data is too large. I wonder if there is a method that can use less data to get a good score.",
    "3188206": "What does this mean? Is this some benchmarking you did yourself? Or data extracted from the paper? \n\nJust my personal opinion, but posts without any context, just dumped data, don't really inspire discussions.",
    "3195167": "I'm wondering how you transpose the dedicated trained models for different styles of velocity map to the unlabelled test data ? ",
    "3197819": ""
  }
}