The baseline model I've included is an Inception Resnet V2 classifier trained on the iNaturalist Idaho and CCT images simultaneously, with no class rebalancing or weighting. This model was a full image classifier, not using the detector. Let me know if you're curious about any additional details!
", "messages": [ { "id": "501703", "postDate": "03/27/2019 16:33:29", "content": "The baseline model I've included is an Inception Resnet V2 classifier trained on the iNaturalist Idaho and CCT images simultaneously, with no class rebalancing or weighting. This model was a full image classifier, not using the detector. Let me know if you're curious about any additional details!
", "rawMarkdown": "The baseline model I've included is an Inception Resnet V2 classifier trained on the iNaturalist Idaho and CCT images simultaneously, with no class rebalancing or weighting. This model was a full image classifier, not using the detector. Let me know if you're curious about any additional details!", "votes": null }, { "id": "502411", "postDate": "03/28/2019 14:53:55", "content": "Thank you for providing a baseline! Did you rescale the image to a smaller size? E.g. 32x32 or 224x224. If so, how did you deal with the memory usage (I think that loading the entire training tensor would require more than just 14 GB of RAM, and using methods like fit_generator is extremely slow)?
", "rawMarkdown": "Thank you for providing a baseline! Did you rescale the image to a smaller size? E.g. 32x32 or 224x224. If so, how did you deal with the memory usage (I think that loading the entire training tensor would require more than just 14 GB of RAM, and using methods like fit_generator is extremely slow)?", "votes": null }, { "id": "502502", "postDate": "03/28/2019 16:53:47", "content": "I used an input size of 299x299, and trained on an AWS instance with 2 NVIDIA Tesla V100 GPUs. I used tensorflow to train, and used the tensorflow tfrecord input pipeline to batch-load the images.
", "rawMarkdown": "I used an input size of 299x299, and trained on an AWS instance with 2 NVIDIA Tesla V100 GPUs. I used tensorflow to train, and used the tensorflow tfrecord input pipeline to batch-load the images.", "votes": null }, { "id": "503566", "postDate": "03/30/2019 08:06:42", "content": "Thank you for your information. I have some confusions.\n- The CCT images are Smaller CCT images provided at Github?. Are they also training data provided at Kaggle too?. \n- Similarly, The test images provided by Kaggle are Smaller IDFG images provided at Github?
Am I correct?
", "rawMarkdown": "Thank you for your information. I have some confusions.\n- The CCT images are `Smaller CCT images` provided at Github?. Are they also `training data` provided at Kaggle too?. \n- Similarly, The test images provided by Kaggle are `Smaller IDFG images` provided at Github? \n\nAm I correct?", "votes": null }, { "id": "503751", "postDate": "03/30/2019 13:34:38", "content": "That's right! If you want to download the full-size images you can do that through the github. Sorry for the confusion!
", "rawMarkdown": "That's right! If you want to download the full-size images you can do that through the github. Sorry for the confusion!", "votes": null }, { "id": "515723", "postDate": "04/13/2019 02:49:48", "content": "Where can I find your baseline model? Thanks!
", "rawMarkdown": "Where can I find your baseline model? Thanks!", "votes": null }, { "id": "522554", "postDate": "04/24/2019 16:28:35", "content": "I'm not providing it, but it shouldn't be hard to replicate!
", "rawMarkdown": "I'm not providing it, but it shouldn't be hard to replicate!", "votes": null }, { "id": "522630", "postDate": "04/24/2019 18:40:03", "content": "Is the baseline score posted somewhere? Sometimes in kaggle competitions its posted on the leaderboard.
", "rawMarkdown": "Is the baseline score posted somewhere? Sometimes in kaggle competitions its posted on the leaderboard.", "votes": null }, { "id": "522636", "postDate": "04/24/2019 18:50:45", "content": "Yes, it's the baseline on the leaderboard.
", "rawMarkdown": "Yes, it's the baseline on the leaderboard.", "votes": null }, { "id": "522637", "postDate": "04/24/2019 18:51:22", "content": "Currently below the 75th team
", "rawMarkdown": "Currently below the 75th team", "votes": null }, { "id": "522716", "postDate": "04/24/2019 22:05:39", "content": "Thanks, that was my fault. Didn’t think it would be that far down.
", "rawMarkdown": "Thanks, that was my fault. Didn’t think it would be that far down.", "votes": null }, { "id": "522725", "postDate": "04/24/2019 22:34:16", "content": "It's not a very creative baseline :)
", "rawMarkdown": "It's not a very creative baseline :)", "votes": null } ], "comments": [ { "id": 502411, "author_name": "xhlulu", "author_url": "", "post_date": "03/28/2019 14:53:55", "content": "Thank you for providing a baseline! Did you rescale the image to a smaller size? E.g. 32x32 or 224x224. If so, how did you deal with the memory usage (I think that loading the entire training tensor would require more than just 14 GB of RAM, and using methods like fit_generator is extremely slow)?
", "votes": null, "replies": [] }, { "id": 502502, "author_name": "sbeery", "author_url": "", "post_date": "03/28/2019 16:53:47", "content": "I used an input size of 299x299, and trained on an AWS instance with 2 NVIDIA Tesla V100 GPUs. I used tensorflow to train, and used the tensorflow tfrecord input pipeline to batch-load the images.
", "votes": null, "replies": [] }, { "id": 503566, "author_name": "backaggle", "author_url": "", "post_date": "03/30/2019 08:06:42", "content": "Thank you for your information. I have some confusions.\n- The CCT images are Smaller CCT images provided at Github?. Are they also training data provided at Kaggle too?. \n- Similarly, The test images provided by Kaggle are Smaller IDFG images provided at Github?
Am I correct?
", "votes": null, "replies": [ { "id": 503751, "author_name": "sbeery", "author_url": "", "post_date": "03/30/2019 13:34:38", "content": "That's right! If you want to download the full-size images you can do that through the github. Sorry for the confusion!
", "votes": null, "replies": [] } ] }, { "id": 515723, "author_name": "joebenet", "author_url": "", "post_date": "04/13/2019 02:49:48", "content": "Where can I find your baseline model? Thanks!
", "votes": null, "replies": [ { "id": 522554, "author_name": "sbeery", "author_url": "", "post_date": "04/24/2019 16:28:35", "content": "I'm not providing it, but it shouldn't be hard to replicate!
", "votes": null, "replies": [] }, { "id": 522630, "author_name": "benweinstein", "author_url": "", "post_date": "04/24/2019 18:40:03", "content": "Is the baseline score posted somewhere? Sometimes in kaggle competitions its posted on the leaderboard.
", "votes": null, "replies": [] }, { "id": 522636, "author_name": "sbeery", "author_url": "", "post_date": "04/24/2019 18:50:45", "content": "Yes, it's the baseline on the leaderboard.
", "votes": null, "replies": [] }, { "id": 522637, "author_name": "sbeery", "author_url": "", "post_date": "04/24/2019 18:51:22", "content": "Currently below the 75th team
", "votes": null, "replies": [] }, { "id": 522716, "author_name": "benweinstein", "author_url": "", "post_date": "04/24/2019 22:05:39", "content": "Thanks, that was my fault. Didn’t think it would be that far down.
", "votes": null, "replies": [] }, { "id": 522725, "author_name": "sbeery", "author_url": "", "post_date": "04/24/2019 22:34:16", "content": "It's not a very creative baseline :)
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "501703": "The baseline model I've included is an Inception Resnet V2 classifier trained on the iNaturalist Idaho and CCT images simultaneously, with no class rebalancing or weighting. This model was a full image classifier, not using the detector. Let me know if you're curious about any additional details!", "502411": "Thank you for providing a baseline! Did you rescale the image to a smaller size? E.g. 32x32 or 224x224. If so, how did you deal with the memory usage (I think that loading the entire training tensor would require more than just 14 GB of RAM, and using methods like fit_generator is extremely slow)?", "502502": "I used an input size of 299x299, and trained on an AWS instance with 2 NVIDIA Tesla V100 GPUs. I used tensorflow to train, and used the tensorflow tfrecord input pipeline to batch-load the images.", "503566": "Thank you for your information. I have some confusions.\n- The CCT images are `Smaller CCT images` provided at Github?. Are they also `training data` provided at Kaggle too?. \n- Similarly, The test images provided by Kaggle are `Smaller IDFG images` provided at Github? \n\nAm I correct?", "503751": "That's right! If you want to download the full-size images you can do that through the github. Sorry for the confusion!", "515723": "Where can I find your baseline model? Thanks!", "522554": "I'm not providing it, but it shouldn't be hard to replicate!", "522630": "Is the baseline score posted somewhere? Sometimes in kaggle competitions its posted on the leaderboard.", "522636": "Yes, it's the baseline on the leaderboard.", "522637": "Currently below the 75th team", "522716": "Thanks, that was my fault. Didn’t think it would be that far down.", "522725": "It's not a very creative baseline :)" }, "source": "meta" }