{
  "id": 547063,
  "title": "Synthetic Data",
  "url": "/competitions/czii-cryo-et-object-identification/discussion/547063",
  "author_name": "",
  "post_date": "2024-11-19T15:33:30.362895500Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi everyone,</p>\n<p>How are you going about generating synthetic data? I know <a href=\"https://www.kaggle.com/hengck23\" target=\"_blank\">@hengck23</a> was talking about it earlier. I was thinking of use generative modeling (VAE/GAN/Diffusion) but was wondering if there are any more simple approaches to this.  </p>\n<p>Also, is the simulated data okay to use for model training?</p>\n<p>Thanks</p>",
  "messages": [
    {
      "id": "3049886",
      "postDate": "11/19/2024 15:33:30",
      "content": "<p>Hi everyone,</p>\n<p>How are you going about generating synthetic data? I know <a href=\"https://www.kaggle.com/hengck23\" target=\"_blank\">@hengck23</a> was talking about it earlier. I was thinking of use generative modeling (VAE/GAN/Diffusion) but was wondering if there are any more simple approaches to this.  </p>\n<p>Also, is the simulated data okay to use for model training?</p>\n<p>Thanks</p>",
      "rawMarkdown": "Hi everyone,\n\nHow are you going about generating synthetic data? I know @hengck23 was talking about it earlier. I was thinking of use generative modeling (VAE/GAN/Diffusion) but was wondering if there are any more simple approaches to this.  \n\nAlso, is the simulated data okay to use for model training?\n\nThanks",
      "votes": null
    },
    {
      "id": "3049899",
      "postDate": "11/19/2024 15:44:17",
      "content": "<p><a href=\"https://www.kaggle.com/circuitguru\" target=\"_blank\">@circuitguru</a> just to point out that we have some precomputed synthetic data here: <a href=\"https://cryoetdataportal.czscience.com/datasets/10441?deposition-id=10310\" target=\"_blank\">https://cryoetdataportal.czscience.com/datasets/10441?deposition-id=10310</a></p>\n<p>we also have some code to generate your own with the PolNet library: <a href=\"https://github.com/copick/copick-catalog/blob/main/solutions/polnet/generate-copick-project/solution.py\" target=\"_blank\">https://github.com/copick/copick-catalog/blob/main/solutions/polnet/generate-copick-project/solution.py</a></p>\n<p>This definitely isn't the only way to generate synthetic data, and there may be some other approaches that work as well or better.</p>\n<p>Cheers!</p>",
      "rawMarkdown": "circuitguru just to point out that we have some precomputed synthetic data here: https://cryoetdataportal.czscience.com/datasets/10441?deposition-id=10310\n\nwe also have some code to generate your own with the PolNet library: https://github.com/copick/copick-catalog/blob/main/solutions/polnet/generate-copick-project/solution.py\n\nThis definitely isn't the only way to generate synthetic data, and there may be some other approaches that work as well or better.\n\nCheers!",
      "votes": null
    },
    {
      "id": "3049943",
      "postDate": "11/19/2024 16:40:04",
      "content": "<p>cheapman's GAN is simply cut --&gt; augment (e.g. rotate) --&gt; paste</p>\n<p>[1] Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation<br>\n<a href=\"https://openaccess.thecvf.com/content/CVPR2021/papers/Ghiasi_Simple_Copy-Paste_Is_a_Strong_Data_Augmentation_Method_for_Instance_CVPR_2021_paper.pdf\" target=\"_blank\">https://openaccess.thecvf.com/content/CVPR2021/papers/Ghiasi_Simple_Copy-Paste_Is_a_Strong_Data_Augmentation_Method_for_Instance_CVPR_2021_paper.pdf</a></p>\n<p>[2] Soft-CP: A Credible and Effective Data Augmentation for Semantic Segmentation<br>\nof Medical Lesions<br>\n<a href=\"https://arxiv.org/pdf/2203.10507\" target=\"_blank\">https://arxiv.org/pdf/2203.10507</a></p>\n<hr>\n<p>if you model is fast enough, \"cut and paste\" can also be use as TTA (test-time augment)</p>",
      "rawMarkdown": "cheapman's GAN is simply cut --> augment (e.g. rotate) --> paste\n\n[1] Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation\nhttps://openaccess.thecvf.com/content/CVPR2021/papers/Ghiasi_Simple_Copy-Paste_Is_a_Strong_Data_Augmentation_Method_for_Instance_CVPR_2021_paper.pdf\n\n\n[2] Soft-CP: A Credible and Effective Data Augmentation for Semantic Segmentation\nof Medical Lesions\nhttps://arxiv.org/pdf/2203.10507\n\n----\n\nif you model is fast enough, \"cut and paste\" can also be use as TTA (test-time augment)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3049899,
      "author_name": "kharrington",
      "author_url": "",
      "post_date": "11/19/2024 15:44:17",
      "content": "<p><a href=\"https://www.kaggle.com/circuitguru\" target=\"_blank\">@circuitguru</a> just to point out that we have some precomputed synthetic data here: <a href=\"https://cryoetdataportal.czscience.com/datasets/10441?deposition-id=10310\" target=\"_blank\">https://cryoetdataportal.czscience.com/datasets/10441?deposition-id=10310</a></p>\n<p>we also have some code to generate your own with the PolNet library: <a href=\"https://github.com/copick/copick-catalog/blob/main/solutions/polnet/generate-copick-project/solution.py\" target=\"_blank\">https://github.com/copick/copick-catalog/blob/main/solutions/polnet/generate-copick-project/solution.py</a></p>\n<p>This definitely isn't the only way to generate synthetic data, and there may be some other approaches that work as well or better.</p>\n<p>Cheers!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3049943,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "11/19/2024 16:40:04",
      "content": "<p>cheapman's GAN is simply cut --&gt; augment (e.g. rotate) --&gt; paste</p>\n<p>[1] Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation<br>\n<a href=\"https://openaccess.thecvf.com/content/CVPR2021/papers/Ghiasi_Simple_Copy-Paste_Is_a_Strong_Data_Augmentation_Method_for_Instance_CVPR_2021_paper.pdf\" target=\"_blank\">https://openaccess.thecvf.com/content/CVPR2021/papers/Ghiasi_Simple_Copy-Paste_Is_a_Strong_Data_Augmentation_Method_for_Instance_CVPR_2021_paper.pdf</a></p>\n<p>[2] Soft-CP: A Credible and Effective Data Augmentation for Semantic Segmentation<br>\nof Medical Lesions<br>\n<a href=\"https://arxiv.org/pdf/2203.10507\" target=\"_blank\">https://arxiv.org/pdf/2203.10507</a></p>\n<hr>\n<p>if you model is fast enough, \"cut and paste\" can also be use as TTA (test-time augment)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3049886": "Hi everyone,\n\nHow are you going about generating synthetic data? I know @hengck23 was talking about it earlier. I was thinking of use generative modeling (VAE/GAN/Diffusion) but was wondering if there are any more simple approaches to this.  \n\nAlso, is the simulated data okay to use for model training?\n\nThanks",
    "3049899": "circuitguru just to point out that we have some precomputed synthetic data here: https://cryoetdataportal.czscience.com/datasets/10441?deposition-id=10310\n\nwe also have some code to generate your own with the PolNet library: https://github.com/copick/copick-catalog/blob/main/solutions/polnet/generate-copick-project/solution.py\n\nThis definitely isn't the only way to generate synthetic data, and there may be some other approaches that work as well or better.\n\nCheers!",
    "3049943": "cheapman's GAN is simply cut --> augment (e.g. rotate) --> paste\n\n[1] Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation\nhttps://openaccess.thecvf.com/content/CVPR2021/papers/Ghiasi_Simple_Copy-Paste_Is_a_Strong_Data_Augmentation_Method_for_Instance_CVPR_2021_paper.pdf\n\n\n[2] Soft-CP: A Credible and Effective Data Augmentation for Semantic Segmentation\nof Medical Lesions\nhttps://arxiv.org/pdf/2203.10507\n\n----\n\nif you model is fast enough, \"cut and paste\" can also be use as TTA (test-time augment)"
  },
  "source": "meta"
}