I use a segmentation approach.. the area of the LV is segmented and calculated for each sax for a given time, then it is combined to get the total volume.
", "messages": [ { "id": "106516", "postDate": "02/01/2016 17:25:49", "content": "I use a segmentation approach.. the area of the LV is segmented and calculated for each sax for a given time, then it is combined to get the total volume.
", "rawMarkdown": "I use a segmentation approach.. the area of the LV is segmented and calculated for each sax for a given time, then it is combined to get the total volume.", "votes": null }, { "id": "106539", "postDate": "02/01/2016 22:53:26", "content": "I’m not <0.018 but just out of curiosity how did you do the segmentation? Did you follow the official deep learning approach using the SunnyBrook dataset? How would this approach be invariant to scale from sample to sample? Your score is impressive so something is obviously working. Are you using Deconvolutional Networks for segmentation (there are no contour labels for this dataset)?
", "rawMarkdown": "I’m not <0.018 but just out of curiosity how did you do the segmentation? Did you follow the official deep learning approach using the SunnyBrook dataset? How would this approach be invariant to scale from sample to sample? Your score is impressive so something is obviously working. Are you using Deconvolutional Networks for segmentation (there are no contour labels for this dataset)?", "votes": null }, { "id": "106540", "postDate": "02/01/2016 22:58:08", "content": "My approach is more like the other tutorial with no neural nets. and I did not use the SunnyBrook dataset... It is naturally invariant to scale. It mimics how a human determines the volume, 1) find the center of the LV 2) find the best contour curve and calculate the area 3) calculate the volume
\n\n[quote=DavidGbodiOdaibo;106539]
\n\nI’m not <0.018 but just out of curiosity how did you do the segmentation? Did you follow the official deep learning approach using the SunnyBrook dataset? How would this approach be invariant to scale from sample to sample? Your score is impressive so something is obviously working.
\n\n[/quote]
", "rawMarkdown": "My approach is more like the other tutorial with no neural nets. and I did not use the SunnyBrook dataset... It is naturally invariant to scale. It mimics how a human determines the volume, 1) find the center of the LV 2) find the best contour curve and calculate the area 3) calculate the volume \r\n\r\n[quote=DavidGbodiOdaibo;106539]\r\n\r\nI’m not <0.018 but just out of curiosity how did you do the segmentation? Did you follow the official deep learning approach using the SunnyBrook dataset? How would this approach be invariant to scale from sample to sample? Your score is impressive so something is obviously working.\r\n\r\n[/quote]", "votes": null }, { "id": "106543", "postDate": "02/01/2016 23:25:49", "content": "One can probably use your approach to find tighter bounding boxes around the heart and then get better and smaller crops of the training images for a neural net. It will eliminate a lot of the none essential details and help the neural net focus better.
", "rawMarkdown": "One can probably use your approach to find tighter bounding boxes around the heart and then get better and smaller crops of the training images for a neural net. It will eliminate a lot of the none essential details and help the neural net focus better.", "votes": null }, { "id": "106546", "postDate": "02/01/2016 23:47:23", "content": "Hi. I might be interested (I'm currently #7). I am using CNNs right now. How can we discuss this further? Is there private messaging on kaggle? I think we have good complementary approaches. However I'd like to find out a little bit more about you and how we might work together before deciding if this merge will be a good idea. Hope that's ok?
", "rawMarkdown": "Hi. I might be interested (I'm currently #7). I am using CNNs right now. How can we discuss this further? Is there private messaging on kaggle? I think we have good complementary approaches. However I'd like to find out a little bit more about you and how we might work together before deciding if this merge will be a good idea. Hope that's ok?", "votes": null }, { "id": "106549", "postDate": "02/01/2016 23:57:47", "content": "@woshialex, That's very impressive. I didn't think that approach would be that competitive, but I'm happy to be proven wrong. That approach seems likely to be more useful approach in practice than the end-to-end approach since it also generates a segmentation.
\n\nI'm not (quite) ready to team up; I'm still hoping I can get past my current road block. However, if I was, this is the kind of model I'd be looking for since it's completely different from what I'm doing.
", "rawMarkdown": "woshialex, That's very impressive. I didn't think that approach would be that competitive, but I'm happy to be proven wrong. That approach seems likely to be more useful approach in practice than the end-to-end approach since it also generates a segmentation. \r\n\r\nI'm not (quite) ready to team up; I'm still hoping I can get past my current road block. However, if I was, this is the kind of model I'd be looking for since it's completely different from what I'm doing.", "votes": null }, { "id": "106553", "postDate": "02/02/2016 00:50:24", "content": "I'd be interested if you want to lower your standards a bit. Or wait until I reach them (hopefully).
", "rawMarkdown": "I'd be interested if you want to lower your standards a bit. Or wait until I reach them (hopefully).", "votes": null } ], "comments": [ { "id": 106539, "author_name": "godaibo", "author_url": "", "post_date": "02/01/2016 22:53:26", "content": "I’m not <0.018 but just out of curiosity how did you do the segmentation? Did you follow the official deep learning approach using the SunnyBrook dataset? How would this approach be invariant to scale from sample to sample? Your score is impressive so something is obviously working. Are you using Deconvolutional Networks for segmentation (there are no contour labels for this dataset)?
", "votes": null, "replies": [] }, { "id": 106540, "author_name": "woshialex", "author_url": "", "post_date": "02/01/2016 22:58:08", "content": "My approach is more like the other tutorial with no neural nets. and I did not use the SunnyBrook dataset... It is naturally invariant to scale. It mimics how a human determines the volume, 1) find the center of the LV 2) find the best contour curve and calculate the area 3) calculate the volume
\n\n[quote=DavidGbodiOdaibo;106539]
\n\nI’m not <0.018 but just out of curiosity how did you do the segmentation? Did you follow the official deep learning approach using the SunnyBrook dataset? How would this approach be invariant to scale from sample to sample? Your score is impressive so something is obviously working.
\n\n[/quote]
", "votes": null, "replies": [] }, { "id": 106543, "author_name": "godaibo", "author_url": "", "post_date": "02/01/2016 23:25:49", "content": "One can probably use your approach to find tighter bounding boxes around the heart and then get better and smaller crops of the training images for a neural net. It will eliminate a lot of the none essential details and help the neural net focus better.
", "votes": null, "replies": [] }, { "id": 106546, "author_name": "tencia", "author_url": "", "post_date": "02/01/2016 23:47:23", "content": "Hi. I might be interested (I'm currently #7). I am using CNNs right now. How can we discuss this further? Is there private messaging on kaggle? I think we have good complementary approaches. However I'd like to find out a little bit more about you and how we might work together before deciding if this merge will be a good idea. Hope that's ok?
", "votes": null, "replies": [] }, { "id": 106549, "author_name": "bitsofbits", "author_url": "", "post_date": "02/01/2016 23:57:47", "content": "@woshialex, That's very impressive. I didn't think that approach would be that competitive, but I'm happy to be proven wrong. That approach seems likely to be more useful approach in practice than the end-to-end approach since it also generates a segmentation.
\n\nI'm not (quite) ready to team up; I'm still hoping I can get past my current road block. However, if I was, this is the kind of model I'd be looking for since it's completely different from what I'm doing.
", "votes": null, "replies": [] }, { "id": 106553, "author_name": "halfwit", "author_url": "", "post_date": "02/02/2016 00:50:24", "content": "I'd be interested if you want to lower your standards a bit. Or wait until I reach them (hopefully).
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "106516": "I use a segmentation approach.. the area of the LV is segmented and calculated for each sax for a given time, then it is combined to get the total volume.", "106539": "I’m not <0.018 but just out of curiosity how did you do the segmentation? Did you follow the official deep learning approach using the SunnyBrook dataset? How would this approach be invariant to scale from sample to sample? Your score is impressive so something is obviously working. Are you using Deconvolutional Networks for segmentation (there are no contour labels for this dataset)?", "106540": "My approach is more like the other tutorial with no neural nets. and I did not use the SunnyBrook dataset... It is naturally invariant to scale. It mimics how a human determines the volume, 1) find the center of the LV 2) find the best contour curve and calculate the area 3) calculate the volume \r\n\r\n[quote=DavidGbodiOdaibo;106539]\r\n\r\nI’m not <0.018 but just out of curiosity how did you do the segmentation? Did you follow the official deep learning approach using the SunnyBrook dataset? How would this approach be invariant to scale from sample to sample? Your score is impressive so something is obviously working.\r\n\r\n[/quote]", "106543": "One can probably use your approach to find tighter bounding boxes around the heart and then get better and smaller crops of the training images for a neural net. It will eliminate a lot of the none essential details and help the neural net focus better.", "106546": "Hi. I might be interested (I'm currently #7). I am using CNNs right now. How can we discuss this further? Is there private messaging on kaggle? I think we have good complementary approaches. However I'd like to find out a little bit more about you and how we might work together before deciding if this merge will be a good idea. Hope that's ok?", "106549": "woshialex, That's very impressive. I didn't think that approach would be that competitive, but I'm happy to be proven wrong. That approach seems likely to be more useful approach in practice than the end-to-end approach since it also generates a segmentation. \r\n\r\nI'm not (quite) ready to team up; I'm still hoping I can get past my current road block. However, if I was, this is the kind of model I'd be looking for since it's completely different from what I'm doing.", "106553": "I'd be interested if you want to lower your standards a bit. Or wait until I reach them (hopefully)." }, "source": "meta" }