I think i might make use of this dataset: https://nihcc.app.box.com/v/ChestXray-NIHCC
", "messages": [ { "id": 376992, "postDate": "2018-08-28T12:44:32.290Z", "content": "I think i might make use of this dataset: https://nihcc.app.box.com/v/ChestXray-NIHCC
", "rawMarkdown": "I think i might make use of this dataset: https://nihcc.app.box.com/v/ChestXray-NIHCC", "votes": 22 }, { "id": 391998, "postDate": "2018-09-22T20:17:39.387Z", "content": "I think using ANY external dataset is highly problematic and should not be permitted at all. If you find some data on the web, how do you know that (1) it is not subset of or did not come for the NIH, (2) was a source FOR the NIH dataset? I have looked through the paper and its not clear where the data came from. There is nothing in the paper saying that the data was collected exclusively for ChestX-ray8. Who knows where the data could have leaked out to from the ORIGINAL sources.
", "rawMarkdown": "I think using ANY external dataset is highly problematic and should not be permitted at all. If you find some data on the web, how do you know that (1) it is not subset of or did not come for the NIH, (2) was a source FOR the NIH dataset? I have looked through the paper and its not clear where the data came from. There is nothing in the paper saying that the data was collected exclusively for ChestX-ray8. Who knows where the data could have leaked out to from the ORIGINAL sources.", "votes": 6 }, { "id": 377414, "postDate": "2018-08-29T06:02:41.517Z", "content": "I think that the images of this competition are already taken from the larger NIH dataset. We also thought about using the above dataset to get more training images. The problem is that we tried to match the images so as not to end up with duplicate images when training and also in order to be able to preserve a validation set. So far we haven't been successful. It would help if the organizers can match the images in this competition's dataset with the corresponding ones in the NIH dataset.
", "rawMarkdown": "I think that the images of this competition are already taken from the larger NIH dataset. We also thought about using the above dataset to get more training images. The problem is that we tried to match the images so as not to end up with duplicate images when training and also in order to be able to preserve a validation set. So far we haven't been successful. It would help if the organizers can match the images in this competition's dataset with the corresponding ones in the NIH dataset.", "votes": 3, "replies": [ { "id": 377855, "postDate": "2018-08-29T21:07:14.923Z", "content": "Sure, that would help. But i thought i could use them as is, not to train the detector, but to train a Backend from scratch. My intuition is, that a network pretrained on Imagenet isnt quite the best lung-expert ,-)\nAnd for that task, it does not matter that much.
", "rawMarkdown": "Sure, that would help. But i thought i could use them as is, not to train the detector, but to train a Backend from scratch. My intuition is, that a network pretrained on Imagenet isnt quite the best lung-expert ,-)\nAnd for that task, it does not matter that much.", "votes": 1 }, { "id": 378490, "postDate": "2018-08-30T09:10:49.260Z", "content": "NIH Chest X-ray dataset has only 1431 pneumonia images, number of patients is even less. So I do not think this dataset is extracted from NIH.
", "rawMarkdown": "NIH Chest X-ray dataset has only 1431 pneumonia images, number of patients is even less. So I do not think this dataset is extracted from NIH.", "votes": 1 }, { "id": 389148, "postDate": "2018-09-18T08:27:22.690Z", "content": "I found just 120 examples of pneumonia by this link: https://nihcc.app.box.com/v/ChestXray-NIHCC Did I understand something wrong?
", "rawMarkdown": "I found just 120 examples of pneumonia by this link: [https://nihcc.app.box.com/v/ChestXray-NIHCC][1] Did I understand something wrong?\n\n\n [1]: https://nihcc.app.box.com/v/ChestXray-NIHCC" } ] }, { "id": 377142, "postDate": "2018-08-28T16:29:24.867Z", "content": "Please consider this the official external data thread, per the competition rules:
\n\n\n", "rawMarkdown": "Please consider this the **official external data thread**, [per the competition rules][1]: \n\n> Publicly, freely available external data is permitted. The source of\n> any external data must be posted to the official competition forum\n> prior to the First Submission Deadline. Entrants may re-annotate\n> images in the training set, but may **not** hand-label predictions,\n> including having human observers rate and evaluate the test data set.\n\n[1]: https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/rules", "votes": 3, "replies": [ { "id": 396569, "postDate": "2018-10-01T02:39:11.300Z", "content": "Publicly, freely available external data is permitted. The source of\n any external data must be posted to the official competition forum\n prior to the First Submission Deadline. Entrants may re-annotate\n images in the training set, but may not hand-label predictions,\n including having human observers rate and evaluate the test data set.
\n
Hi Julia, what does it mean by \"freely available external data\"? Can I use NIH PLCO Chest X-ray dataset? It is a public dataset and everyone can apply for the data without paying.
", "rawMarkdown": "Hi Julia, what does it mean by \"freely available external data\"? Can I use NIH PLCO Chest X-ray dataset? It is a public dataset and everyone can apply for the data without paying." } ] }, { "id": 379778, "postDate": "2018-09-01T00:00:18.297Z", "content": "https://github.com/arnoweng/CheXNet
", "rawMarkdown": "https://github.com/arnoweng/CheXNet", "votes": 1, "replies": [ { "id": 379780, "postDate": "2018-09-01T00:06:57.017Z", "content": "I'm going to toy with the pre-trainied model above.
", "rawMarkdown": "I'm going to toy with the pre-trainied model above." }, { "id": 379864, "postDate": "2018-09-01T05:07:42.063Z", "content": "0.78 is not as good as this ;-)
", "rawMarkdown": "0.78 is not as good as [this][1] ;-)\n\n\n [1]: https://github.com/Azure/AzureChestXRay" } ] }, { "id": 378616, "postDate": "2018-08-30T12:18:09.577Z", "content": "All the pertained models from pytroch 0.41
", "rawMarkdown": "All the pertained models from pytroch 0.41 ", "votes": 1 }, { "id": 404956, "postDate": "2018-10-16T16:37:19.937Z", "content": "We are using https://github.com/experiencor/keras-yolo3 with associated pre-trained weights, as well as pre-trained weights found in https://github.com/brucechou1983/CheXNet-Keras.
", "rawMarkdown": "We are using https://github.com/experiencor/keras-yolo3 with associated pre-trained weights, as well as pre-trained weights found in https://github.com/brucechou1983/CheXNet-Keras.", "votes": 2 }, { "id": 397430, "postDate": "2018-10-02T13:25:09.700Z", "content": "https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5
\n\nas in https://www.kaggle.com/hmendonca/mask-rcnn-and-coco-transfer-learning-lb-0-155
", "rawMarkdown": "https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n\nas in https://www.kaggle.com/hmendonca/mask-rcnn-and-coco-transfer-learning-lb-0-155" }, { "id": 397117, "postDate": "2018-10-01T22:46:03.823Z", "content": "The competition organizers have discussed some of the recurring questions on this thread. Here are their clarifications:\n1. You may use the NIH labels.\n2. You may use the CheXnet models.
", "rawMarkdown": "The competition organizers have discussed some of the recurring questions on this thread. Here are their clarifications:\n1. You may use the NIH labels.\n2. You may use the CheXnet models.", "replies": [ { "id": 397148, "postDate": "2018-10-02T01:24:22.290Z", "content": "\n\n\nUsing the NIH dataset and associated annotations outside of the\n training set is not permitted. Other than that exception, publicly,\n freely available external data is permitted.
\n
Hi Julia,
\n\nSorry, I don't mean to be rude, but why would you want to change that 2 weeks to the merger deadline?\nI believe allowing the additional dataset and state of the art models would be good. However, at this point in time, it is quite a change of scope for the competition and would require additional time for the participants to analyze the data and adapt all their models.
\n\nThanks for your understanding and all the time spent organizing these competitions!
", "rawMarkdown": "> Using the NIH dataset and associated annotations outside of the\n> training set is not permitted. Other than that exception, publicly,\n> freely available external data is permitted.\n\nHi Julia,\n\nSorry, I don't mean to be rude, but why would you want to change that 2 weeks to the merger deadline?\nI believe allowing the additional dataset and state of the art models would be good. However, at this point in time, it is quite a change of scope for the competition and would require additional time for the participants to analyze the data and adapt all their models.\n\nThanks for your understanding and all the time spent organizing these competitions!", "votes": 7 }, { "id": 397155, "postDate": "2018-10-02T01:48:46.490Z", "content": "I agree with Henrique. I think the organizers and everybody will get more out of this if we focus on using the data provided --- only --- and develop a good approach to building a model to solve this problem. Later on, after the contest is over, the model could be improved by adding more data from whatever other sources are available.
\n\nFurthermore, changing the rules of the contest at this point is really unfair for those of us who have until now followed the rules strictly; for us its really a completely different contest now.
", "rawMarkdown": "I agree with Henrique. I think the organizers and everybody will get more out of this if we focus on using the data provided --- only --- and develop a good approach to building a model to solve this problem. Later on, after the contest is over, the model could be improved by adding more data from whatever other sources are available.\n\nFurthermore, changing the rules of the contest at this point is really unfair for those of us who have until now followed the rules strictly; for us its really a completely different contest now.", "votes": 2 }, { "id": 397184, "postDate": "2018-10-02T03:40:55.907Z", "content": "Hi Julia,
\n\nThanks for clarifications. Please let me confirm. Using NIH labels is permitted only for training purpose. Is this right?
", "rawMarkdown": "Hi Julia,\n\nThanks for clarifications. Please let me confirm. Using NIH labels is permitted only for training purpose. Is this right?", "votes": 1 }, { "id": 397251, "postDate": "2018-10-02T07:30:10.157Z", "content": "I want to clarify.\nNIH dataset can be separated in this way.
\n\n#1 NIH images inside RSNA training data
\n#2 NIH images inside RSNA test1 data
\n#3 NIH images inside RSNA test2 data
\n#4 NIH images outside RSNA dataset
\n#5 NIH tabular data inside RSNA training data
\n#6 NIH tabular data inside RSNA test1 data
\n#7 NIH tabular data inside RSNA test2 data
\n#8 NIH tabular data outside RSNA dataset
\n#9 NIH bounding box data inside RSNA training data
\n#10 NIH bounding box data inside RSNA test1 data
\n#11 NIH bounding box data inside RSNA test2 data
\n#12 NIH bounding box data outside RSNA dataset
Now, in my understanding, all of #1 - #12 is permitted for training.\nIs it correct?
", "rawMarkdown": "I want to clarify.\nNIH dataset can be separated in this way.\n\n\\#1 NIH images inside RSNA training data \n\\#2 NIH images inside RSNA test1 data \n\\#3 NIH images inside RSNA test2 data \n\\#4 NIH images outside RSNA dataset \n\\#5 NIH tabular data inside RSNA training data \n\\#6 NIH tabular data inside RSNA test1 data \n\\#7 NIH tabular data inside RSNA test2 data \n\\#8 NIH tabular data outside RSNA dataset \n\\#9 NIH bounding box data inside RSNA training data \n\\#10 NIH bounding box data inside RSNA test1 data \n\\#11 NIH bounding box data inside RSNA test2 data \n\\#12 NIH bounding box data outside RSNA dataset \n\nNow, in my understanding, all of #1 - #12 is permitted for training.\nIs it correct?", "votes": 2 }, { "id": 399444, "postDate": "2018-10-05T19:39:26.760Z", "content": "In response to your clarification question, all 12 of those groups of data may be used. However with the reminder that you are not permitted to hand annotate / add new annotations to anything outside of the images that have been provided in the RSNA training data (#1 per OsciiArt's list above). So the images themselves and dataset as it is publicly available today is permitted for use, but hand annotations on top of those non-training set images are not permitted.
", "rawMarkdown": "In response to your clarification question, all 12 of those groups of data may be used. However with the reminder that you are not permitted to hand annotate / add new annotations to anything outside of the images that have been provided in the RSNA training data (#1 per OsciiArt's list above). So the images themselves and dataset as it is publicly available today is permitted for use, but hand annotations on top of those non-training set images are not permitted.", "votes": 2 } ] }, { "id": 409834, "postDate": "2018-10-24T23:03:59.453Z", "content": "https://github.com/arnoweng/CheXNet
", "rawMarkdown": "https://github.com/arnoweng/CheXNet" }, { "id": 409776, "postDate": "2018-10-24T20:23:29.677Z", "content": "Pytorch has pretrained models available, see the links in these files https://github.com/pytorch/vision/tree/master/torchvision/models
", "rawMarkdown": "Pytorch has pretrained models available, see the links in these files https://github.com/pytorch/vision/tree/master/torchvision/models" }, { "id": 409711, "postDate": "2018-10-24T18:25:56.113Z", "content": "Starting weights from: https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5
\n\nAnd also starting script : https://www.kaggle.com/hmendonca/mask-rcnn-and-coco-transfer-learning-lb-0-155
", "rawMarkdown": "Starting weights from: https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n\nAnd also starting script : https://www.kaggle.com/hmendonca/mask-rcnn-and-coco-transfer-learning-lb-0-155" }, { "id": 409666, "postDate": "2018-10-24T17:04:08.503Z", "content": "Using https://github.com/open-mmlab/mmdetection with pre-trained weights on imagenet, as well as https://github.com/matterport/Mask_RCNN with pre-trained weights on coco
", "rawMarkdown": "Using https://github.com/open-mmlab/mmdetection with pre-trained weights on imagenet, as well as https://github.com/matterport/Mask_RCNN with pre-trained weights on coco" }, { "id": 409498, "postDate": "2018-10-24T11:49:27.560Z", "content": "NIH data：\nhttps://www.kaggle.com/nih-chest-xrays/data\npretrain model：\nhttps://github.com/matterport/Mask_RCNN/releases\nhttps://github.com/MachineLP/models/tree/master/research/slim
", "rawMarkdown": "NIH data：\nhttps://www.kaggle.com/nih-chest-xrays/data\npretrain model：\nhttps://github.com/matterport/Mask_RCNN/releases\nhttps://github.com/MachineLP/models/tree/master/research/slim\n" }, { "id": 409337, "postDate": "2018-10-24T06:19:00.183Z", "content": "For us we have been using\nExternal Data:\nhttps://www.kaggle.com/nih-chest-xrays/data\nOpensource code and pretrained weight:\nhttps://github.com/qqwweee/keras-yolo3\nhttps://github.com/matterport/Mask_RCNN\nhttps://github.com/tensorflow/models/tree/master/research/object_detection\nhttps://github.com/arnoweng/CheXNet\nhttps://github.com/imlab-uiip/lung-segmentation-2d
\n\nHope everyone enjoy the competition. Good luck with the stage 2!
", "rawMarkdown": "For us we have been using\nExternal Data:\nhttps://www.kaggle.com/nih-chest-xrays/data\nOpensource code and pretrained weight:\nhttps://github.com/qqwweee/keras-yolo3\nhttps://github.com/matterport/Mask_RCNN\nhttps://github.com/tensorflow/models/tree/master/research/object_detection\nhttps://github.com/arnoweng/CheXNet\nhttps://github.com/imlab-uiip/lung-segmentation-2d\n\nHope everyone enjoy the competition. Good luck with the stage 2!" }, { "id": 409291, "postDate": "2018-10-24T04:13:27.207Z", "content": "Using keras-retinanet and mask-rcnn (pretrained on coco)
", "rawMarkdown": "Using keras-retinanet and mask-rcnn (pretrained on coco)" }, { "id": 409275, "postDate": "2018-10-24T02:59:04.163Z", "content": "We are using \nMaskrcnn\nchest_xray\ndeeplabv3+
", "rawMarkdown": "We are using \n[Maskrcnn][1]\n[chest_xray][2]\n[deeplabv3+][3]\n\n [1]: https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n [2]: https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia\n [3]: https://github.com/bonlime/keras-deeplab-v3-plus" }, { "id": 409244, "postDate": "2018-10-24T01:54:17.747Z", "content": "we are using mask_rcnn_coco.h5 in MaskrRCNN\nwe are using weights in deeplabv3+
", "rawMarkdown": "[we are using mask_rcnn_coco.h5 in MaskrRCNN][1]\n[we are using weights in deeplabv3+][2]\n\n\n [1]: https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n [2]: https://github.com/bonlime/keras-deeplab-v3-plus" }, { "id": 409242, "postDate": "2018-10-24T01:49:13.660Z", "content": "https://github.com/bonlime/keras-deeplab-v3-plus
", "rawMarkdown": "https://github.com/bonlime/keras-deeplab-v3-plus" }, { "id": 408985, "postDate": "2018-10-23T17:32:02.863Z", "content": "All the pretrained models from https://github.com/qubvel/segmentation_models
", "rawMarkdown": "All the pretrained models from https://github.com/qubvel/segmentation_models" }, { "id": 408539, "postDate": "2018-10-23T03:37:50.730Z", "content": "We are using https://github.com/matterport/Mask_RCNN and https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5 and\nhttps://github.com/zoogzog/chexnet
", "rawMarkdown": "We are using [https://github.com/matterport/Mask_RCNN][1] and [https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5][2] and\n[https://github.com/zoogzog/chexnet][3]\n\n\n [1]: https://github.com/matterport/Mask_RCNN\n [2]: https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n [3]: https://github.com/zoogzog/chexnet" }, { "id": 407518, "postDate": "2018-10-21T11:37:14.373Z", "content": "Great idea.
", "rawMarkdown": "Great idea." }, { "id": 406519, "postDate": "2018-10-19T11:35:04.787Z", "content": "If it hasn't already been mentionned: https://github.com/fchollet/deep-learning-models/releases/download/v0.2/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5
", "rawMarkdown": "If it hasn't already been mentionned: https://github.com/fchollet/deep-learning-models/releases/download/v0.2/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5" }, { "id": 406335, "postDate": "2018-10-19T04:25:37.040Z", "content": "I am using this https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5
", "rawMarkdown": "I am using this https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5" }, { "id": 406326, "postDate": "2018-10-19T04:00:46.733Z", "content": "We've used the following pre-trained models:
\n\nhttp://download.tensorflow.org/models/object_detection/ssd_mobilenet_v2_coco_2018_03_29.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_inception_v2_coco_2018_01_28.tar.gz\nhttps://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_resnet50_coco_2018_01_28.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_resnet101_coco_2018_01_28.tar.gz\nhttps://pjreddie.com/media/files/darknet53.conv.74\nhttp://download.tensorflow.org/models/resnet_v1_152_2016_08_28.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_nas_coco_2018_01_28.tar.gz
", "rawMarkdown": "We've used the following pre-trained models:\n\nhttp://download.tensorflow.org/models/object_detection/ssd_mobilenet_v2_coco_2018_03_29.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_inception_v2_coco_2018_01_28.tar.gz\nhttps://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_resnet50_coco_2018_01_28.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_resnet101_coco_2018_01_28.tar.gz\nhttps://pjreddie.com/media/files/darknet53.conv.74\nhttp://download.tensorflow.org/models/resnet_v1_152_2016_08_28.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_nas_coco_2018_01_28.tar.gz" }, { "id": 406096, "postDate": "2018-10-18T16:22:06.253Z", "content": "There are a few chest dataset available here on kaggle:
\n\nPretrained models:
\n\nChainerCV pretrained models https://github.com/chainer/chainercv (links to weight files are available in code, e.g., https://github.com/chainer/chainercv/blob/d8f903acbc6e3369a9871aa175893fa5c0f3946c/chainercv/links/model/resnet/resnet.py#L115-L116)
\n\ndeep-residual-networks https://github.com/KaimingHe/deep-residual-networks
\n\npretrained-models.pytorch https://github.com/Cadene/pretrained-models.pytorch
\n\nSENet https://github.com/hujie-frank/SENet
\n\nData:
\n\nNIH Chest X-rays https://www.kaggle.com/nih-chest-xrays/data
", "rawMarkdown": "Pretrained models:\n\nChainerCV pretrained models https://github.com/chainer/chainercv (links to weight files are available in code, e.g., https://github.com/chainer/chainercv/blob/d8f903acbc6e3369a9871aa175893fa5c0f3946c/chainercv/links/model/resnet/resnet.py#L115-L116)\n\ndeep-residual-networks https://github.com/KaimingHe/deep-residual-networks\n\npretrained-models.pytorch https://github.com/Cadene/pretrained-models.pytorch\n\nSENet https://github.com/hujie-frank/SENet\n\nData:\n\nNIH Chest X-rays https://www.kaggle.com/nih-chest-xrays/data" }, { "id": 405686, "postDate": "2018-10-17T23:06:30.430Z", "content": "Adding another pre-trained CheXNet model:
\n\nhttps://github.com/zoogzog/chexnet/tree/master/models
", "rawMarkdown": "Adding another pre-trained CheXNet model:\n\nhttps://github.com/zoogzog/chexnet/tree/master/models" }, { "id": 405575, "postDate": "2018-10-17T18:19:39.793Z", "content": "It looks like I need a PAID account to be able to file a ticket to figure out why I cant download that dataset. Do PAID for datasets qualify as \"publicly available\"?
", "rawMarkdown": "It looks like I need a PAID account to be able to file a ticket to figure out why I cant download that dataset. Do PAID for datasets qualify as \"publicly available\"?", "replies": [ { "id": 405591, "postDate": "2018-10-17T18:59:56.270Z", "content": "OK, I will answer my own question. That dataset does not qualify because, as per competition rules, its not \"freely available\".
", "rawMarkdown": "OK, I will answer my own question. That dataset does not qualify because, as per competition rules, its not \"freely available\"." }, { "id": 405597, "postDate": "2018-10-17T19:22:00.173Z", "content": "Not sure why you can't download the NIH dataset. I just go to the link, click on the file, and hit Download without having to sign in or do anything else.
", "rawMarkdown": "Not sure why you can't download the NIH dataset. I just go to the link, click on the file, and hit Download without having to sign in or do anything else." }, { "id": 405630, "postDate": "2018-10-17T20:35:10.193Z", "content": "I downloaded it this evening with no problems..
", "rawMarkdown": "I downloaded it this evening with no problems.. " }, { "id": 405671, "postDate": "2018-10-17T22:17:01.727Z", "content": "I succeeded in downloading the first tar file. The rest I cannot download. I do not know why but Box.com wants me to pay them before they support my questions.
", "rawMarkdown": "I succeeded in downloading the first tar file. The rest I cannot download. I do not know why but Box.com wants me to pay them before they support my questions." }, { "id": 405673, "postDate": "2018-10-17T22:23:00.717Z", "content": "You can try here https://www.kaggle.com/nih-chest-xrays/data
", "rawMarkdown": "You can try here https://www.kaggle.com/nih-chest-xrays/data" }, { "id": 405683, "postDate": "2018-10-17T22:49:28.060Z", "content": "I think it would be better to be able to download that dataset using the kaggle api, oh, you just posted that link...
", "rawMarkdown": "I think it would be better to be able to download that dataset using the kaggle api, oh, you just posted that link..." } ] }, { "id": 405488, "postDate": "2018-10-17T15:25:04.297Z", "content": "https://github.com/tonylins/pytorch-mobilenet-v2
", "rawMarkdown": "https://github.com/tonylins/pytorch-mobilenet-v2" }, { "id": 405314, "postDate": "2018-10-17T08:16:16.013Z", "content": "Pretrained model from:\n - https://keras.io/applications/\n - https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n - https://pjreddie.com/media/files/darknet53.conv.74\n - https://github.com/fizyr/keras-retinanet/releases/download/0.4.1/resnet50_coco_best_v2.1.0.h5
", "rawMarkdown": "Pretrained model from:\n - https://keras.io/applications/\n - https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n - https://pjreddie.com/media/files/darknet53.conv.74\n - https://github.com/fizyr/keras-retinanet/releases/download/0.4.1/resnet50_coco_best_v2.1.0.h5" }, { "id": 405173, "postDate": "2018-10-17T02:13:12.897Z", "content": "pretrained models from \nhttps://github.com/Cadene/pretrained-models.pytorch \nhttps://github.com/guoruoqian/cascade-rcnn_Pytorch
", "rawMarkdown": "pretrained models from \nhttps://github.com/Cadene/pretrained-models.pytorch \nhttps://github.com/guoruoqian/cascade-rcnn_Pytorch" }, { "id": 404908, "postDate": "2018-10-16T14:52:25.187Z", "content": "We use pre-trained models from https://github.com/Cadene/pretrained-models.pytorch and https://github.com/pytorch/vision/tree/master/torchvision/models.
", "rawMarkdown": "We use pre-trained models from https://github.com/Cadene/pretrained-models.pytorch and https://github.com/pytorch/vision/tree/master/torchvision/models." }, { "id": 404599, "postDate": "2018-10-16T04:01:42.663Z", "content": "https://github.com/bonlime/keras-deeplab-v3-plus
", "rawMarkdown": "https://github.com/bonlime/keras-deeplab-v3-plus" }, { "id": 403982, "postDate": "2018-10-15T03:11:01.883Z", "content": "ImageNet Pretrained Models:
\n\nWe are using NIH datasets from\nhttps://nihcc.app.box.com/v/ChestXray-NIHCC/folder/37178474737
", "rawMarkdown": "We are using NIH datasets from\nhttps://nihcc.app.box.com/v/ChestXray-NIHCC/folder/37178474737", "replies": [ { "id": 405567, "postDate": "2018-10-17T18:05:52.803Z", "content": "I cannot download that data. It seems to require a special account with Box.com. So it does not seem publicly available.
", "rawMarkdown": "I cannot download that data. It seems to require a special account with Box.com. So it does not seem publicly available." } ] }, { "id": 402887, "postDate": "2018-10-12T14:15:51.897Z", "content": "Pretrained RetinaNet models:\nhttps://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018
", "rawMarkdown": "Pretrained RetinaNet models:\nhttps://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018" }, { "id": 402545, "postDate": "2018-10-11T21:20:18.987Z", "content": "Pre-trained models available in
\n\nNIH chest xray dataset\nhttps://nihcc.app.box.com/v/ChestXray-NIHCC
\n\nPretrained models from \nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://keras.io/ja/applications/\nhttps://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md
", "rawMarkdown": "NIH chest xray dataset\nhttps://nihcc.app.box.com/v/ChestXray-NIHCC\n\nPretrained models from \nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://keras.io/ja/applications/\nhttps://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md" }, { "id": 401860, "postDate": "2018-10-10T19:38:03.373Z", "content": "I'm using pretrained models from https://github.com/Cadene/pretrained-models.pytorch
", "rawMarkdown": "I'm using pretrained models from https://github.com/Cadene/pretrained-models.pytorch" }, { "id": 400526, "postDate": "2018-10-08T13:13:49.193Z", "content": "pre-trained models in TensorFlow detection model zoo (https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md).
", "rawMarkdown": "pre-trained models in TensorFlow detection model zoo (https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md)." }, { "id": 400129, "postDate": "2018-10-07T16:48:17.920Z", "content": "https://github.com/YuwenXiong/py-R-FCN and associated pre-trained models.
", "rawMarkdown": "https://github.com/YuwenXiong/py-R-FCN and associated pre-trained models." }, { "id": 396844, "postDate": "2018-10-01T13:26:46.900Z", "content": "Anything from https://github.com/facebookresearch/Detectron and https://github.com/roytseng-tw/Detectron.pytorch
", "rawMarkdown": "Anything from [https://github.com/facebookresearch/Detectron][1] and https://github.com/roytseng-tw/Detectron.pytorch\n\n\n [1]: https://github.com/facebookresearch/Detectron" }, { "id": 396777, "postDate": "2018-10-01T11:08:31.700Z", "content": "ResNet-101 ImageNet pre-trained: https://github.com/fizyr/keras-models/releases/download/v0.0.1/ResNet-101-model.keras.h5
\n\nResNet-152 ImageNet pre-trained: https://github.com/fizyr/keras-models/releases/download/v0.0.1/ResNet-152-model.keras.h5
\n\nResNet-v1-101 ImageNet pre-trained: https://1drv.ms/u/s!Am-5JzdW2XHzhqMEtxf1Ciym8uZ8sg
", "rawMarkdown": "ResNet-101 ImageNet pre-trained: https://github.com/fizyr/keras-models/releases/download/v0.0.1/ResNet-101-model.keras.h5\n\nResNet-152 ImageNet pre-trained: https://github.com/fizyr/keras-models/releases/download/v0.0.1/ResNet-152-model.keras.h5\n\nResNet-v1-101 ImageNet pre-trained: https://1drv.ms/u/s!Am-5JzdW2XHzhqMEtxf1Ciym8uZ8sg", "replies": [ { "id": 403172, "postDate": "2018-10-13T02:39:20.270Z", "content": "Along the same lines:
\n\nResNet-50 ImageNet pre-trained: https://github.com/fizyr/keras-models/releases/download/v0.0.1/ResNet-50-model.keras.h5
", "rawMarkdown": "Along the same lines:\n\nResNet-50 ImageNet pre-trained: [https://github.com/fizyr/keras-models/releases/download/v0.0.1/ResNet-50-model.keras.h5][1]\n\n\n [1]: https://github.com/fizyr/keras-models/releases/download/v0.0.1/ResNet-50-model.keras.h5" } ] }, { "id": 394457, "postDate": "2018-09-26T21:48:09.327Z", "content": "Very interested in the teams' discussion on the variability of the data sets. I mean, what did you expect? The disease is not 'easy' to 'pin down' which is why the challenge was set, was it not? There is so much discussion about the techniques to 'bound' the targets, but has anyone considered looking at the bone structures (for example), or any other clues from the available data? I mean skeletons (bone structures) tell stories... so many teams seem focused on just one aspect, how about considering the available data on the health of each patient?
", "rawMarkdown": "Very interested in the teams' discussion on the variability of the data sets. I mean, what did you expect? The disease is not 'easy' to 'pin down' which is why the challenge was set, was it not? There is so much discussion about the techniques to 'bound' the targets, but has anyone considered looking at the bone structures (for example), or any other clues from the available data? I mean skeletons (bone structures) tell stories... so many teams seem focused on just one aspect, how about considering the available data on the health of each patient?", "replies": [ { "id": 394515, "postDate": "2018-09-27T02:20:08.650Z", "content": "We expected training set with roughly the same characteristics as the teet set. Pinning down pneumonia is difficult enough without artificial obstecales. To give an example, what you said is like saying that you shouldn't complain if your driving lessons are all at midnight. Its not that its easier then driving at noon, just very different.
", "rawMarkdown": "We expected training set with roughly the same characteristics as the teet set. Pinning down pneumonia is difficult enough without artificial obstecales. To give an example, what you said is like saying that you shouldn't complain if your driving lessons are all at midnight. Its not that its easier then driving at noon, just very different. " } ] }, { "id": 389888, "postDate": "2018-09-19T11:56:10.697Z", "content": "Using the NIH dataset and associated annotations outside of the training set is not permitted. Can Somebody explain the above line. I am little confused. Does it mean that I can only use images provided by the competitors and no other images from NIH for training purpose
", "rawMarkdown": "Using the NIH dataset and associated annotations outside of the training set is not permitted. Can Somebody explain the above line. I am little confused. Does it mean that I can only use images provided by the competitors and no other images from NIH for training purpose" }, { "id": 389135, "postDate": "2018-09-18T07:59:56.593Z", "content": "I found just 120 examples with pneumonia in this dataset: https://nihcc.app.box.com/v/ChestXray-NIHCC\nDid I understand something wrong?
", "rawMarkdown": "I found just 120 examples with pneumonia in this dataset: [https://nihcc.app.box.com/v/ChestXray-NIHCC][1]\nDid I understand something wrong?\n\n [1]: https://nihcc.app.box.com/v/ChestXray-NIHCC", "replies": [ { "id": 389160, "postDate": "2018-09-18T09:21:33.117Z", "content": "This is a very small subset of the database that also has bbox. The big database only has multiclass categories
", "rawMarkdown": "This is a very small subset of the database that also has bbox. The big database only has multiclass categories" } ] }, { "id": 387559, "postDate": "2018-09-15T07:02:42.457Z", "content": "https://github.com/fchollet/deep-learning-models/releases
", "rawMarkdown": "https://github.com/fchollet/deep-learning-models/releases" }, { "id": 382058, "postDate": "2018-09-05T16:21:02.797Z", "content": "https://data.mendeley.com/datasets/rscbjbr9sj/3
", "rawMarkdown": "https://data.mendeley.com/datasets/rscbjbr9sj/3" }, { "id": 381789, "postDate": "2018-09-05T07:45:26.943Z", "content": "https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia
", "rawMarkdown": "https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia" }, { "id": 381510, "postDate": "2018-09-04T18:10:12Z", "content": "https://www.kaggle.com/nih-chest-xrays/data
", "rawMarkdown": "https://www.kaggle.com/nih-chest-xrays/data" }, { "id": 380064, "postDate": "2018-09-01T15:24:00.210Z", "content": "keras pre-trained models\nhttps://keras.io/applications/
", "rawMarkdown": "keras pre-trained models\nhttps://keras.io/applications/" }, { "id": 378885, "postDate": "2018-08-30T16:28:03.237Z", "content": "The question has come up around use of other annotated datasets that are based off of the same set of underlying chest x-ray images from the NIH. First, a clarification that these other public datasets annotated on the NIH images are not the labels used in any part of this competition's provided train or test sets. The host team independently labeled the images for this competition specifically. In addition, as per the competition's rules, \"Entrants may re-annotate images in the training set, but may not hand-label predictions, including having human observers rate and evaluate the test data set.\" This means that any hand annotations - including external data which has been annotated as such - are only permitted to be used on the training set of images. Solutions that make use of any form of annotation on the original NIH images outside of the training set may be grounds for disqualification.
", "rawMarkdown": "The question has come up around use of other annotated datasets that are based off of the same set of underlying chest x-ray images from the NIH. First, a clarification that these other public datasets annotated on the NIH images are **not** the labels used in any part of this competition's provided train or test sets. The host team independently labeled the images for this competition specifically. In addition, as per the competition's rules, \"Entrants may re-annotate images in the training set, but may not hand-label predictions, including having human observers rate and evaluate the test data set.\" This means that any hand annotations - including external data which has been annotated as such - are only permitted to be used on the **training set** of images. Solutions that make use of any form of annotation on the original NIH images outside of the training set may be grounds for disqualification. ", "replies": [ { "id": 379246, "postDate": "2018-08-31T04:30:18.583Z", "content": "So, are we allowed to use NIH for training purposes, giving the fact that \"test\" images are part of it and are already labeled?
\n\nKeep in mind that using NIH will probably overfit the classification problem, but would obviously not solve the segmentation one.
", "rawMarkdown": "So, are we allowed to use NIH for training purposes, giving the fact that \"test\" images are part of it and are already labeled? \n\nKeep in mind that using NIH will probably overfit the classification problem, but would obviously not solve the segmentation one.", "votes": 4 }, { "id": 382570, "postDate": "2018-09-06T16:47:22.133Z", "content": "Please re-read my message above.
\n\n\n", "rawMarkdown": "Please re-read my message above. \n\n\n> This means that any hand annotations - including external data which\n> has been annotated as such - are only permitted to be used on the\n> training set of images. Solutions that make use of any form of\n> annotation on the original NIH images outside of the training set may\n> be grounds for disqualification." }, { "id": 387907, "postDate": "2018-09-15T22:30:50.437Z", "content": "This means that any hand annotations - including external data which\n has been annotated as such - are only permitted to be used on the\n training set of images. Solutions that make use of any form of\n annotation on the original NIH images outside of the training set may\n be grounds for disqualification.
\n
I want to use class (findings) data of NIH (Chest X-ray 14) data. \nIt may be beneficial to pre-train a model by NIH images and the labels. But if I do so, this model must memorize which class the images included in stage 2 belong to. I can not avoid this because I don't know which images of NIH become stage2 images. Is this permitted?
", "rawMarkdown": "I want to use class (findings) data of NIH (Chest X-ray 14) data. \nIt may be beneficial to pre-train a model by NIH images and the labels. But if I do so, this model must memorize which class the images included in stage 2 belong to. I can not avoid this because I don't know which images of NIH become stage2 images. Is this permitted?", "votes": 4 }, { "id": 388857, "postDate": "2018-09-17T17:52:46.840Z", "content": "After reading all the comments here I'm still not confident you are allowed to use NIH data for training or not. I appreciate if anyone clarify this.
\n\n\n\n\nPublicly, freely available external data is permitted.
\n
This means using NIH data is permitted I assume.
\n\n\n\n\nSolutions that make use of any form of annotation on the original NIH\n images outside of the training set may be grounds for\n disqualification.
\n
So using NIH data is not permitted?
", "rawMarkdown": "After reading all the comments here I'm still not confident you are allowed to use NIH data for training or not. I appreciate if anyone clarify this.\n\n> Publicly, freely available external data is permitted.\n\nThis means using NIH data is permitted I assume.\n\n\n> Solutions that make use of any form of annotation on the original NIH\n> images outside of the training set may be grounds for\n> disqualification.\n\nSo using NIH data is not permitted?", "votes": 2 }, { "id": 388962, "postDate": "2018-09-17T22:46:27.560Z", "content": "Using the NIH dataset and associated annotations outside of the training set is not permitted. Other than that exception, publicly, freely available external data is permitted.
", "rawMarkdown": "Using the NIH dataset and associated annotations outside of the training set is not permitted. Other than that exception, publicly, freely available external data is permitted." }, { "id": 388986, "postDate": "2018-09-17T23:59:49.347Z", "content": "Sigh, this is still a bit unclear. Ok, let's be specific. Suppose i use chexnet as part of my ensamble. I didn't use nih for training it, but the model has been trained on nih data. Is that permitted?
", "rawMarkdown": "Sigh, this is still a bit unclear. Ok, let's be specific. Suppose i use chexnet as part of my ensamble. I didn't use nih for training it, but the model has been trained on nih data. Is that permitted? ", "votes": 5 }, { "id": 389044, "postDate": "2018-09-18T03:47:42.057Z", "content": "Thank you for a quick answer Julia. Now I understand more clear.
\n\n\n\n\nSigh, this is still a bit unclear. Ok, let's be specific. Suppose i\n use chexnet as part of my ensamble. I didn't use nih for training it,\n but the model has been trained on nih data. Is that permitted?
\n
According to my understanding, that is not permitted because chexnet is trained on NIH data many of which are outside of the provided training set. Hope I'm correct.
", "rawMarkdown": "Thank you for a quick answer Julia. Now I understand more clear.\n\n> Sigh, this is still a bit unclear. Ok, let's be specific. Suppose i\n> use chexnet as part of my ensamble. I didn't use nih for training it,\n> but the model has been trained on nih data. Is that permitted?\n\nAccording to my understanding, that is not permitted because chexnet is trained on NIH data many of which are outside of the provided training set. Hope I'm correct.", "votes": 1 }, { "id": 389064, "postDate": "2018-09-18T05:03:49.190Z", "content": "I would like to hear this from the organizers.
", "rawMarkdown": "I would like to hear this from the organizers.", "votes": 2 }, { "id": 389338, "postDate": "2018-09-18T15:03:02.633Z", "content": "@Julia Thank you for your patience and kind reply. I think I understand. Umm... It's a bit disappointing. If I could use them, I could try weakly-supervised learning using image-level annotations. And also I could try semi-supervised learning using just images. It should be fun.
", "rawMarkdown": "@Julia Thank you for your patience and kind reply. I think I understand. Umm... It's a bit disappointing. If I could use them, I could try weakly-supervised learning using image-level annotations. And also I could try semi-supervised learning using just images. It should be fun." }, { "id": 395321, "postDate": "2018-09-28T10:13:25.777Z", "content": "According to the rules it is not permitted to use the test set for training. I guess this is the reason why you wrote that it is forbidden to use any image in NIH dataset outside the training set. \nHowever, after the stage 2 images will be published it will be very easy to remove all test images from the NIH data. Is it permitted to optimize the models using the NIH dataset after we remove from it the testset? \nJust to be clear - I mean that there will be no part of the model that uses the test set - only the complete NIH dataset WITHOUT the testset.
", "rawMarkdown": "According to the rules it is not permitted to use the test set for training. I guess this is the reason why you wrote that it is forbidden to use any image in NIH dataset outside the training set. \nHowever, after the stage 2 images will be published it will be very easy to remove all test images from the NIH data. Is it permitted to optimize the models using the NIH dataset after we remove from it the testset? \nJust to be clear - I mean that there will be no part of the model that uses the test set - only the complete NIH dataset WITHOUT the testset." } ] }, { "id": 377772, "postDate": "2018-08-29T18:11:23.460Z", "content": "Are there requirements for posting pre-trained models?
", "rawMarkdown": "Are there requirements for posting pre-trained models?", "replies": [ { "id": 377790, "postDate": "2018-08-29T19:00:41Z", "content": "You should default to doing so, in the event that what you consider to be pre-trained models is considered external data by the host.
", "rawMarkdown": "You should default to doing so, in the event that what you consider to be pre-trained models is considered external data by the host." } ] }, { "id": 406333, "postDate": "2018-10-19T04:18:35.577Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 407021, "postDate": "2018-10-20T07:58:50.247Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 397423, "postDate": "2018-10-02T13:14:56.390Z", "content": "https://www.kaggle.com/kmader/pulmonary-chest-xray-abnormalities
", "rawMarkdown": "https://www.kaggle.com/kmader/pulmonary-chest-xray-abnormalities", "isDeleted": true }, { "id": 379223, "postDate": "2018-08-31T03:25:06.433Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 391998, "author_name": "snapperwiffer", "author_url": "", "post_date": "2018-09-22T20:17:39.387000", "content": "I think using ANY external dataset is highly problematic and should not be permitted at all. If you find some data on the web, how do you know that (1) it is not subset of or did not come for the NIH, (2) was a source FOR the NIH dataset? I have looked through the paper and its not clear where the data came from. There is nothing in the paper saying that the data was collected exclusively for ChestX-ray8. Who knows where the data could have leaked out to from the ORIGINAL sources.
", "votes": 6, "replies": [] }, { "id": 377414, "author_name": "GuyE", "author_url": "", "post_date": "2018-08-29T06:02:41.517000", "content": "I think that the images of this competition are already taken from the larger NIH dataset. We also thought about using the above dataset to get more training images. The problem is that we tried to match the images so as not to end up with duplicate images when training and also in order to be able to preserve a validation set. So far we haven't been successful. It would help if the organizers can match the images in this competition's dataset with the corresponding ones in the NIH dataset.
", "votes": 3, "replies": [ { "id": 377855, "author_name": "Michael Schimmer", "author_url": "", "post_date": "2018-08-29T21:07:14.923000", "content": "Sure, that would help. But i thought i could use them as is, not to train the detector, but to train a Backend from scratch. My intuition is, that a network pretrained on Imagenet isnt quite the best lung-expert ,-)\nAnd for that task, it does not matter that much.
", "votes": 1, "replies": [] }, { "id": 378490, "author_name": "Alexander Firsov", "author_url": "", "post_date": "2018-08-30T09:10:49.260000", "content": "NIH Chest X-ray dataset has only 1431 pneumonia images, number of patients is even less. So I do not think this dataset is extracted from NIH.
", "votes": 1, "replies": [] }, { "id": 389148, "author_name": "Anton Makarenko", "author_url": "", "post_date": "2018-09-18T08:27:22.690000", "content": "I found just 120 examples of pneumonia by this link: https://nihcc.app.box.com/v/ChestXray-NIHCC Did I understand something wrong?
", "votes": 0, "replies": [] } ] }, { "id": 377142, "author_name": "Julia Elliott", "author_url": "", "post_date": "2018-08-28T16:29:24.867000", "content": "Please consider this the official external data thread, per the competition rules:
\n\n\n", "votes": 3, "replies": [ { "id": 396569, "author_name": "Allen", "author_url": "", "post_date": "2018-10-01T02:39:11.300000", "content": "Publicly, freely available external data is permitted. The source of\n any external data must be posted to the official competition forum\n prior to the First Submission Deadline. Entrants may re-annotate\n images in the training set, but may not hand-label predictions,\n including having human observers rate and evaluate the test data set.
\n
Hi Julia, what does it mean by \"freely available external data\"? Can I use NIH PLCO Chest X-ray dataset? It is a public dataset and everyone can apply for the data without paying.
", "votes": 0, "replies": [] } ] }, { "id": 379778, "author_name": "Bryan Arnold", "author_url": "", "post_date": "2018-09-01T00:00:18.297000", "content": "https://github.com/arnoweng/CheXNet
", "votes": 1, "replies": [ { "id": 379780, "author_name": "Bryan Arnold", "author_url": "", "post_date": "2018-09-01T00:06:57.017000", "content": "I'm going to toy with the pre-trainied model above.
", "votes": 0, "replies": [] }, { "id": 379864, "author_name": "Mihai Cvasnievschi", "author_url": "", "post_date": "2018-09-01T05:07:42.063000", "content": "0.78 is not as good as this ;-)
", "votes": 0, "replies": [] } ] }, { "id": 378616, "author_name": "DrHB", "author_url": "", "post_date": "2018-08-30T12:18:09.577000", "content": "All the pertained models from pytroch 0.41
", "votes": 1, "replies": [] }, { "id": 404956, "author_name": "Ern", "author_url": "", "post_date": "2018-10-16T16:37:19.937000", "content": "We are using https://github.com/experiencor/keras-yolo3 with associated pre-trained weights, as well as pre-trained weights found in https://github.com/brucechou1983/CheXNet-Keras.
", "votes": 2, "replies": [] }, { "id": 397430, "author_name": "Henrique Mendonça", "author_url": "", "post_date": "2018-10-02T13:25:09.700000", "content": "https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5
\n\nas in https://www.kaggle.com/hmendonca/mask-rcnn-and-coco-transfer-learning-lb-0-155
", "votes": 0, "replies": [] }, { "id": 397117, "author_name": "Julia Elliott", "author_url": "", "post_date": "2018-10-01T22:46:03.823000", "content": "The competition organizers have discussed some of the recurring questions on this thread. Here are their clarifications:\n1. You may use the NIH labels.\n2. You may use the CheXnet models.
", "votes": 0, "replies": [ { "id": 397148, "author_name": "Henrique Mendonça", "author_url": "", "post_date": "2018-10-02T01:24:22.290000", "content": "\n\n\nUsing the NIH dataset and associated annotations outside of the\n training set is not permitted. Other than that exception, publicly,\n freely available external data is permitted.
\n
Hi Julia,
\n\nSorry, I don't mean to be rude, but why would you want to change that 2 weeks to the merger deadline?\nI believe allowing the additional dataset and state of the art models would be good. However, at this point in time, it is quite a change of scope for the competition and would require additional time for the participants to analyze the data and adapt all their models.
\n\nThanks for your understanding and all the time spent organizing these competitions!
", "votes": 7, "replies": [] }, { "id": 397155, "author_name": "snapperwiffer", "author_url": "", "post_date": "2018-10-02T01:48:46.490000", "content": "I agree with Henrique. I think the organizers and everybody will get more out of this if we focus on using the data provided --- only --- and develop a good approach to building a model to solve this problem. Later on, after the contest is over, the model could be improved by adding more data from whatever other sources are available.
\n\nFurthermore, changing the rules of the contest at this point is really unfair for those of us who have until now followed the rules strictly; for us its really a completely different contest now.
", "votes": 2, "replies": [] }, { "id": 397184, "author_name": "AkiraSosa", "author_url": "", "post_date": "2018-10-02T03:40:55.907000", "content": "Hi Julia,
\n\nThanks for clarifications. Please let me confirm. Using NIH labels is permitted only for training purpose. Is this right?
", "votes": 1, "replies": [] }, { "id": 397251, "author_name": "OsciiArt", "author_url": "", "post_date": "2018-10-02T07:30:10.157000", "content": "I want to clarify.\nNIH dataset can be separated in this way.
\n\n#1 NIH images inside RSNA training data
\n#2 NIH images inside RSNA test1 data
\n#3 NIH images inside RSNA test2 data
\n#4 NIH images outside RSNA dataset
\n#5 NIH tabular data inside RSNA training data
\n#6 NIH tabular data inside RSNA test1 data
\n#7 NIH tabular data inside RSNA test2 data
\n#8 NIH tabular data outside RSNA dataset
\n#9 NIH bounding box data inside RSNA training data
\n#10 NIH bounding box data inside RSNA test1 data
\n#11 NIH bounding box data inside RSNA test2 data
\n#12 NIH bounding box data outside RSNA dataset
Now, in my understanding, all of #1 - #12 is permitted for training.\nIs it correct?
", "votes": 2, "replies": [] }, { "id": 399444, "author_name": "Julia Elliott", "author_url": "", "post_date": "2018-10-05T19:39:26.760000", "content": "In response to your clarification question, all 12 of those groups of data may be used. However with the reminder that you are not permitted to hand annotate / add new annotations to anything outside of the images that have been provided in the RSNA training data (#1 per OsciiArt's list above). So the images themselves and dataset as it is publicly available today is permitted for use, but hand annotations on top of those non-training set images are not permitted.
", "votes": 2, "replies": [] } ] }, { "id": 409834, "author_name": "Jarrel Seah", "author_url": "", "post_date": "2018-10-24T23:03:59.453000", "content": "https://github.com/arnoweng/CheXNet
", "votes": 0, "replies": [] }, { "id": 409776, "author_name": "Ankoor Bhagat", "author_url": "", "post_date": "2018-10-24T20:23:29.677000", "content": "Pytorch has pretrained models available, see the links in these files https://github.com/pytorch/vision/tree/master/torchvision/models
", "votes": 0, "replies": [] }, { "id": 409711, "author_name": "Alexander", "author_url": "", "post_date": "2018-10-24T18:25:56.113000", "content": "Starting weights from: https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5
\n\nAnd also starting script : https://www.kaggle.com/hmendonca/mask-rcnn-and-coco-transfer-learning-lb-0-155
", "votes": 0, "replies": [] }, { "id": 409666, "author_name": "Xin Yi", "author_url": "", "post_date": "2018-10-24T17:04:08.503000", "content": "Using https://github.com/open-mmlab/mmdetection with pre-trained weights on imagenet, as well as https://github.com/matterport/Mask_RCNN with pre-trained weights on coco
", "votes": 0, "replies": [] }, { "id": 409498, "author_name": "zikun.wei", "author_url": "", "post_date": "2018-10-24T11:49:27.560000", "content": "NIH data：\nhttps://www.kaggle.com/nih-chest-xrays/data\npretrain model：\nhttps://github.com/matterport/Mask_RCNN/releases\nhttps://github.com/MachineLP/models/tree/master/research/slim
", "votes": 0, "replies": [] }, { "id": 409337, "author_name": "Chundi Liu", "author_url": "", "post_date": "2018-10-24T06:19:00.183000", "content": "For us we have been using\nExternal Data:\nhttps://www.kaggle.com/nih-chest-xrays/data\nOpensource code and pretrained weight:\nhttps://github.com/qqwweee/keras-yolo3\nhttps://github.com/matterport/Mask_RCNN\nhttps://github.com/tensorflow/models/tree/master/research/object_detection\nhttps://github.com/arnoweng/CheXNet\nhttps://github.com/imlab-uiip/lung-segmentation-2d
\n\nHope everyone enjoy the competition. Good luck with the stage 2!
", "votes": 0, "replies": [] }, { "id": 409291, "author_name": "Sam Shleifer", "author_url": "", "post_date": "2018-10-24T04:13:27.207000", "content": "Using keras-retinanet and mask-rcnn (pretrained on coco)
", "votes": 0, "replies": [] }, { "id": 409275, "author_name": "dinghuoyijiao", "author_url": "", "post_date": "2018-10-24T02:59:04.163000", "content": "We are using \nMaskrcnn\nchest_xray\ndeeplabv3+
", "votes": 0, "replies": [] }, { "id": 409244, "author_name": "justin000", "author_url": "", "post_date": "2018-10-24T01:54:17.747000", "content": "we are using mask_rcnn_coco.h5 in MaskrRCNN\nwe are using weights in deeplabv3+
", "votes": 0, "replies": [] }, { "id": 409242, "author_name": "justin000", "author_url": "", "post_date": "2018-10-24T01:49:13.660000", "content": "https://github.com/bonlime/keras-deeplab-v3-plus
", "votes": 0, "replies": [] }, { "id": 408985, "author_name": "Virilo Tejedor Aguilera", "author_url": "", "post_date": "2018-10-23T17:32:02.863000", "content": "All the pretrained models from https://github.com/qubvel/segmentation_models
", "votes": 0, "replies": [] }, { "id": 408539, "author_name": "SmartQ", "author_url": "", "post_date": "2018-10-23T03:37:50.730000", "content": "We are using https://github.com/matterport/Mask_RCNN and https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5 and\nhttps://github.com/zoogzog/chexnet
", "votes": 0, "replies": [] }, { "id": 407518, "author_name": "Yee Ng", "author_url": "", "post_date": "2018-10-21T11:37:14.373000", "content": "Great idea.
", "votes": 0, "replies": [] }, { "id": 406519, "author_name": "BaggyG", "author_url": "", "post_date": "2018-10-19T11:35:04.787000", "content": "If it hasn't already been mentionned: https://github.com/fchollet/deep-learning-models/releases/download/v0.2/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5
", "votes": 0, "replies": [] }, { "id": 406335, "author_name": "YaGana Sheriff-Hussaini", "author_url": "", "post_date": "2018-10-19T04:25:37.040000", "content": "I am using this https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5
", "votes": 0, "replies": [] }, { "id": 406326, "author_name": "Peter Yu", "author_url": "", "post_date": "2018-10-19T04:00:46.733000", "content": "We've used the following pre-trained models:
\n\nhttp://download.tensorflow.org/models/object_detection/ssd_mobilenet_v2_coco_2018_03_29.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_inception_v2_coco_2018_01_28.tar.gz\nhttps://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_resnet50_coco_2018_01_28.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_resnet101_coco_2018_01_28.tar.gz\nhttps://pjreddie.com/media/files/darknet53.conv.74\nhttp://download.tensorflow.org/models/resnet_v1_152_2016_08_28.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_nas_coco_2018_01_28.tar.gz
", "votes": 0, "replies": [] }, { "id": 406096, "author_name": "Amil Gentili", "author_url": "", "post_date": "2018-10-18T16:22:06.253000", "content": "There are a few chest dataset available here on kaggle:
\n\nPretrained models:
\n\nChainerCV pretrained models https://github.com/chainer/chainercv (links to weight files are available in code, e.g., https://github.com/chainer/chainercv/blob/d8f903acbc6e3369a9871aa175893fa5c0f3946c/chainercv/links/model/resnet/resnet.py#L115-L116)
\n\ndeep-residual-networks https://github.com/KaimingHe/deep-residual-networks
\n\npretrained-models.pytorch https://github.com/Cadene/pretrained-models.pytorch
\n\nSENet https://github.com/hujie-frank/SENet
\n\nData:
\n\nNIH Chest X-rays https://www.kaggle.com/nih-chest-xrays/data
", "votes": 0, "replies": [] }, { "id": 405686, "author_name": "Epoch of Belief", "author_url": "", "post_date": "2018-10-17T23:06:30.430000", "content": "Adding another pre-trained CheXNet model:
\n\nhttps://github.com/zoogzog/chexnet/tree/master/models
", "votes": 0, "replies": [] }, { "id": 405575, "author_name": "snapperwiffer", "author_url": "", "post_date": "2018-10-17T18:19:39.793000", "content": "It looks like I need a PAID account to be able to file a ticket to figure out why I cant download that dataset. Do PAID for datasets qualify as \"publicly available\"?
", "votes": 0, "replies": [ { "id": 405591, "author_name": "snapperwiffer", "author_url": "", "post_date": "2018-10-17T18:59:56.270000", "content": "OK, I will answer my own question. That dataset does not qualify because, as per competition rules, its not \"freely available\".
", "votes": 0, "replies": [] }, { "id": 405597, "author_name": "Ian Pan", "author_url": "", "post_date": "2018-10-17T19:22:00.173000", "content": "Not sure why you can't download the NIH dataset. I just go to the link, click on the file, and hit Download without having to sign in or do anything else.
", "votes": 0, "replies": [] }, { "id": 405630, "author_name": "Tom Aindow", "author_url": "", "post_date": "2018-10-17T20:35:10.193000", "content": "I downloaded it this evening with no problems..
", "votes": 0, "replies": [] }, { "id": 405671, "author_name": "snapperwiffer", "author_url": "", "post_date": "2018-10-17T22:17:01.727000", "content": "I succeeded in downloading the first tar file. The rest I cannot download. I do not know why but Box.com wants me to pay them before they support my questions.
", "votes": 0, "replies": [] }, { "id": 405673, "author_name": "Tom Aindow", "author_url": "", "post_date": "2018-10-17T22:23:00.717000", "content": "You can try here https://www.kaggle.com/nih-chest-xrays/data
", "votes": 0, "replies": [] }, { "id": 405683, "author_name": "snapperwiffer", "author_url": "", "post_date": "2018-10-17T22:49:28.060000", "content": "I think it would be better to be able to download that dataset using the kaggle api, oh, you just posted that link...
", "votes": 0, "replies": [] } ] }, { "id": 405488, "author_name": "AkiraSosa", "author_url": "", "post_date": "2018-10-17T15:25:04.297000", "content": "", "votes": 0, "replies": [] }, { "id": 405314, "author_name": "", "author_url": "", "post_date": "2018-10-17T08:16:16.013000", "content": "", "votes": 0, "replies": [] }, { "id": 405173, "author_name": "", "author_url": "", "post_date": "2018-10-17T02:13:12.897000", "content": "", "votes": 0, "replies": [] }, { "id": 404908, "author_name": "", "author_url": "", "post_date": "2018-10-16T14:52:25.187000", "content": "", "votes": 0, "replies": [] }, { "id": 404599, "author_name": "", "author_url": "", "post_date": "2018-10-16T04:01:42.663000", "content": "", "votes": 0, "replies": [] }, { "id": 403982, "author_name": "", "author_url": "", "post_date": "2018-10-15T03:11:01.883000", "content": "", "votes": 0, "replies": [] }, { "id": 403666, "author_name": "", "author_url": "", "post_date": "2018-10-14T09:20:22.140000", "content": "", "votes": 0, "replies": [ { "id": 405567, "author_name": "", "author_url": "", "post_date": "2018-10-17T18:05:52.803000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 402887, "author_name": "", "author_url": "", "post_date": "2018-10-12T14:15:51.897000", "content": "", "votes": 0, "replies": [] }, { "id": 402545, "author_name": "", "author_url": "", "post_date": "2018-10-11T21:20:18.987000", "content": "", "votes": 0, "replies": [] }, { "id": 402090, "author_name": "", "author_url": "", "post_date": "2018-10-11T07:05:56.583000", "content": "", "votes": 0, "replies": [] }, { "id": 401860, "author_name": "", "author_url": "", "post_date": "2018-10-10T19:38:03.373000", "content": "", "votes": 0, "replies": [] }, { "id": 400526, "author_name": "", "author_url": "", "post_date": "2018-10-08T13:13:49.193000", "content": "", "votes": 0, "replies": [] }, { "id": 400129, "author_name": "", "author_url": "", "post_date": "2018-10-07T16:48:17.920000", "content": "", "votes": 0, "replies": [] }, { "id": 396844, "author_name": "", "author_url": "", "post_date": "2018-10-01T13:26:46.900000", "content": "", "votes": 0, "replies": [] }, { "id": 396777, "author_name": "", "author_url": "", "post_date": "2018-10-01T11:08:31.700000", "content": "", "votes": 0, "replies": [ { "id": 403172, "author_name": "", "author_url": "", "post_date": "2018-10-13T02:39:20.270000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 394457, "author_name": "", "author_url": "", "post_date": "2018-09-26T21:48:09.327000", "content": "", "votes": 0, "replies": [ { "id": 394515, "author_name": "", "author_url": "", "post_date": "2018-09-27T02:20:08.650000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 389888, "author_name": "", "author_url": "", "post_date": "2018-09-19T11:56:10.697000", "content": "", "votes": 0, "replies": [] }, { "id": 389135, "author_name": "", "author_url": "", "post_date": "2018-09-18T07:59:56.593000", "content": "", "votes": 0, "replies": [ { "id": 389160, "author_name": "", "author_url": "", "post_date": "2018-09-18T09:21:33.117000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 387559, "author_name": "", "author_url": "", "post_date": "2018-09-15T07:02:42.457000", "content": "", "votes": 0, "replies": [] }, { "id": 382058, "author_name": "", "author_url": "", "post_date": "2018-09-05T16:21:02.797000", "content": "", "votes": 0, "replies": [] }, { "id": 381789, "author_name": "", "author_url": "", "post_date": "2018-09-05T07:45:26.943000", "content": "", "votes": 0, "replies": [] }, { "id": 381510, "author_name": "", "author_url": "", "post_date": "2018-09-04T18:10:12", "content": "", "votes": 0, "replies": [] }, { "id": 380064, "author_name": "", "author_url": "", "post_date": "2018-09-01T15:24:00.210000", "content": "", "votes": 0, "replies": [] }, { "id": 378885, "author_name": "", "author_url": "", "post_date": "2018-08-30T16:28:03.237000", "content": "", "votes": 0, "replies": [ { "id": 379246, "author_name": "", "author_url": "", "post_date": "2018-08-31T04:30:18.583000", "content": "", "votes": 4, "replies": [] }, { "id": 382570, "author_name": "", "author_url": "", "post_date": "2018-09-06T16:47:22.133000", "content": "", "votes": 0, "replies": [] }, { "id": 387907, "author_name": "", "author_url": "", "post_date": "2018-09-15T22:30:50.437000", "content": "", "votes": 4, "replies": [] }, { "id": 388857, "author_name": "", "author_url": "", "post_date": "2018-09-17T17:52:46.840000", "content": "", "votes": 2, "replies": [] }, { "id": 388962, "author_name": "", "author_url": "", "post_date": "2018-09-17T22:46:27.560000", "content": "", "votes": 0, "replies": [] }, { "id": 388986, "author_name": "", "author_url": "", "post_date": "2018-09-17T23:59:49.347000", "content": "", "votes": 5, "replies": [] }, { "id": 389044, "author_name": "", "author_url": "", "post_date": "2018-09-18T03:47:42.057000", "content": "", "votes": 1, "replies": [] }, { "id": 389064, "author_name": "", "author_url": "", "post_date": "2018-09-18T05:03:49.190000", "content": "", "votes": 2, "replies": [] }, { "id": 389338, "author_name": "", "author_url": "", "post_date": "2018-09-18T15:03:02.633000", "content": "", "votes": 0, "replies": [] }, { "id": 395321, "author_name": "", "author_url": "", "post_date": "2018-09-28T10:13:25.777000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 377772, "author_name": "", "author_url": "", "post_date": "2018-08-29T18:11:23.460000", "content": "", "votes": 0, "replies": [ { "id": 377790, "author_name": "", "author_url": "", "post_date": "2018-08-29T19:00:41", "content": "", "votes": 0, "replies": [] } ] }, { "id": 406333, "author_name": "", "author_url": "", "post_date": "2018-10-19T04:18:35.577000", "content": "", "votes": 0, "replies": [ { "id": 407021, "author_name": "", "author_url": "", "post_date": "2018-10-20T07:58:50.247000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 397423, "author_name": "", "author_url": "", "post_date": "2018-10-02T13:14:56.390000", "content": "", "votes": 0, "replies": [] }, { "id": 379223, "author_name": "", "author_url": "", "post_date": "2018-08-31T03:25:06.433000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "376992": "I think i might make use of this dataset: https://nihcc.app.box.com/v/ChestXray-NIHCC", "391998": "I think using ANY external dataset is highly problematic and should not be permitted at all. If you find some data on the web, how do you know that (1) it is not subset of or did not come for the NIH, (2) was a source FOR the NIH dataset? I have looked through the paper and its not clear where the data came from. There is nothing in the paper saying that the data was collected exclusively for ChestX-ray8. Who knows where the data could have leaked out to from the ORIGINAL sources.", "377414": "I think that the images of this competition are already taken from the larger NIH dataset. We also thought about using the above dataset to get more training images. The problem is that we tried to match the images so as not to end up with duplicate images when training and also in order to be able to preserve a validation set. So far we haven't been successful. It would help if the organizers can match the images in this competition's dataset with the corresponding ones in the NIH dataset.", "377142": "Please consider this the **official external data thread**, [per the competition rules][1]: \n\n> Publicly, freely available external data is permitted. The source of\n> any external data must be posted to the official competition forum\n> prior to the First Submission Deadline. Entrants may re-annotate\n> images in the training set, but may **not** hand-label predictions,\n> including having human observers rate and evaluate the test data set.\n\n[1]: https://www.kaggle.com/c/rsna-pneumonia-detection-challenge/rules", "379778": "https://github.com/arnoweng/CheXNet", "378616": "All the pertained models from pytroch 0.41 ", "404956": "We are using https://github.com/experiencor/keras-yolo3 with associated pre-trained weights, as well as pre-trained weights found in https://github.com/brucechou1983/CheXNet-Keras.", "397430": "https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n\nas in https://www.kaggle.com/hmendonca/mask-rcnn-and-coco-transfer-learning-lb-0-155", "397117": "The competition organizers have discussed some of the recurring questions on this thread. Here are their clarifications:\n1. You may use the NIH labels.\n2. You may use the CheXnet models.", "409834": "https://github.com/arnoweng/CheXNet", "409776": "Pytorch has pretrained models available, see the links in these files https://github.com/pytorch/vision/tree/master/torchvision/models", "409711": "Starting weights from: https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n\nAnd also starting script : https://www.kaggle.com/hmendonca/mask-rcnn-and-coco-transfer-learning-lb-0-155", "409666": "Using https://github.com/open-mmlab/mmdetection with pre-trained weights on imagenet, as well as https://github.com/matterport/Mask_RCNN with pre-trained weights on coco", "409498": "NIH data：\nhttps://www.kaggle.com/nih-chest-xrays/data\npretrain model：\nhttps://github.com/matterport/Mask_RCNN/releases\nhttps://github.com/MachineLP/models/tree/master/research/slim\n", "409337": "For us we have been using\nExternal Data:\nhttps://www.kaggle.com/nih-chest-xrays/data\nOpensource code and pretrained weight:\nhttps://github.com/qqwweee/keras-yolo3\nhttps://github.com/matterport/Mask_RCNN\nhttps://github.com/tensorflow/models/tree/master/research/object_detection\nhttps://github.com/arnoweng/CheXNet\nhttps://github.com/imlab-uiip/lung-segmentation-2d\n\nHope everyone enjoy the competition. Good luck with the stage 2!", "409291": "Using keras-retinanet and mask-rcnn (pretrained on coco)", "409275": "We are using \n[Maskrcnn][1]\n[chest_xray][2]\n[deeplabv3+][3]\n\n [1]: https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n [2]: https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia\n [3]: https://github.com/bonlime/keras-deeplab-v3-plus", "409244": "[we are using mask_rcnn_coco.h5 in MaskrRCNN][1]\n[we are using weights in deeplabv3+][2]\n\n\n [1]: https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n [2]: https://github.com/bonlime/keras-deeplab-v3-plus", "409242": "https://github.com/bonlime/keras-deeplab-v3-plus", "408985": "All the pretrained models from https://github.com/qubvel/segmentation_models", "408539": "We are using [https://github.com/matterport/Mask_RCNN][1] and [https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5][2] and\n[https://github.com/zoogzog/chexnet][3]\n\n\n [1]: https://github.com/matterport/Mask_RCNN\n [2]: https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n [3]: https://github.com/zoogzog/chexnet", "407518": "Great idea.", "406519": "If it hasn't already been mentionned: https://github.com/fchollet/deep-learning-models/releases/download/v0.2/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5", "406335": "I am using this https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5", "406326": "We've used the following pre-trained models:\n\nhttp://download.tensorflow.org/models/object_detection/ssd_mobilenet_v2_coco_2018_03_29.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_inception_v2_coco_2018_01_28.tar.gz\nhttps://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_resnet50_coco_2018_01_28.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_resnet101_coco_2018_01_28.tar.gz\nhttps://pjreddie.com/media/files/darknet53.conv.74\nhttp://download.tensorflow.org/models/resnet_v1_152_2016_08_28.tar.gz\nhttp://download.tensorflow.org/models/object_detection/faster_rcnn_nas_coco_2018_01_28.tar.gz", "406096": "There are a few chest dataset available here on kaggle:\n\n 1. https://www.kaggle.com/nih-chest-xrays/data \n 2. https://www.kaggle.com/yashprakash13/chest-xrays-dataset \n 3. https://www.kaggle.com/parthachakraborty/pneumonia-chest-x-ray \n 4. https://www.kaggle.com/karan172/chest-xrays \n 5. https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia \n 6. https://www.kaggle.com/kmader/pulmonary-chest-xray-abnormalitie`s", "405836": "Pretrained models:\n\nChainerCV pretrained models https://github.com/chainer/chainercv (links to weight files are available in code, e.g., https://github.com/chainer/chainercv/blob/d8f903acbc6e3369a9871aa175893fa5c0f3946c/chainercv/links/model/resnet/resnet.py#L115-L116)\n\ndeep-residual-networks https://github.com/KaimingHe/deep-residual-networks\n\npretrained-models.pytorch https://github.com/Cadene/pretrained-models.pytorch\n\nSENet https://github.com/hujie-frank/SENet\n\nData:\n\nNIH Chest X-rays https://www.kaggle.com/nih-chest-xrays/data", "405686": "Adding another pre-trained CheXNet model:\n\nhttps://github.com/zoogzog/chexnet/tree/master/models", "405575": "It looks like I need a PAID account to be able to file a ticket to figure out why I cant download that dataset. Do PAID for datasets qualify as \"publicly available\"?", "405488": "https://github.com/tonylins/pytorch-mobilenet-v2", "405314": "Pretrained model from:\n - https://keras.io/applications/\n - https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n - https://pjreddie.com/media/files/darknet53.conv.74\n - https://github.com/fizyr/keras-retinanet/releases/download/0.4.1/resnet50_coco_best_v2.1.0.h5", "405173": "pretrained models from \nhttps://github.com/Cadene/pretrained-models.pytorch \nhttps://github.com/guoruoqian/cascade-rcnn_Pytorch", "404908": "We use pre-trained models from https://github.com/Cadene/pretrained-models.pytorch and https://github.com/pytorch/vision/tree/master/torchvision/models.", "404599": "https://github.com/bonlime/keras-deeplab-v3-plus", "403982": "ImageNet Pretrained Models:\n\n* https://s3-us-west-2.amazonaws.com/detectron/ImageNetPretrained/MSRA/R-50.pkl\n* https://s3-us-west-2.amazonaws.com/detectron/ImageNetPretrained/MSRA/R-101.pkl\n* https://s3-us-west-2.amazonaws.com/detectron/ImageNetPretrained/FBResNeXt/X-101-64x4d.pkl\n* https://s3-us-west-2.amazonaws.com/detectron/ImageNetPretrained/20171220/X-101-32x8d.pkl\n* https://s3-us-west-2.amazonaws.com/detectron/ImageNetPretrained/25093814/X-152-32x8d-IN5k.pkl", "403666": "We are using NIH datasets from\nhttps://nihcc.app.box.com/v/ChestXray-NIHCC/folder/37178474737", "402887": "Pretrained RetinaNet models:\nhttps://github.com/ZFTurbo/Keras-RetinaNet-for-Open-Images-Challenge-2018", "402545": " Pre-trained models available in\n\n 1. torchvision.models in Pytorch\n 2. From this repo: https://github.com/arnoweng/CheXNet\n 3. https://github.com/matterport/Mask_RCNN/releases/download/v2.0/mask_rcnn_coco.h5\n 4. Trained models available in https://www.kaggle.com/hmendonca/mask-rcnn-and-coco-transfer-learning-lb-0-155\n 5. ImageNet pre-trained available in Keras", "402090": "NIH chest xray dataset\nhttps://nihcc.app.box.com/v/ChestXray-NIHCC\n\nPretrained models from \nhttps://github.com/Cadene/pretrained-models.pytorch\nhttps://keras.io/ja/applications/\nhttps://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md", "401860": "I'm using pretrained models from https://github.com/Cadene/pretrained-models.pytorch", "400526": "pre-trained models in TensorFlow detection model zoo (https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md).", "400129": "https://github.com/YuwenXiong/py-R-FCN and associated pre-trained models.", "396844": "Anything from [https://github.com/facebookresearch/Detectron][1] and https://github.com/roytseng-tw/Detectron.pytorch\n\n\n [1]: https://github.com/facebookresearch/Detectron", "396777": "ResNet-101 ImageNet pre-trained: https://github.com/fizyr/keras-models/releases/download/v0.0.1/ResNet-101-model.keras.h5\n\nResNet-152 ImageNet pre-trained: https://github.com/fizyr/keras-models/releases/download/v0.0.1/ResNet-152-model.keras.h5\n\nResNet-v1-101 ImageNet pre-trained: https://1drv.ms/u/s!Am-5JzdW2XHzhqMEtxf1Ciym8uZ8sg", "394457": "Very interested in the teams' discussion on the variability of the data sets. I mean, what did you expect? The disease is not 'easy' to 'pin down' which is why the challenge was set, was it not? There is so much discussion about the techniques to 'bound' the targets, but has anyone considered looking at the bone structures (for example), or any other clues from the available data? I mean skeletons (bone structures) tell stories... so many teams seem focused on just one aspect, how about considering the available data on the health of each patient?", "389888": "Using the NIH dataset and associated annotations outside of the training set is not permitted. Can Somebody explain the above line. I am little confused. Does it mean that I can only use images provided by the competitors and no other images from NIH for training purpose", "389135": "I found just 120 examples with pneumonia in this dataset: [https://nihcc.app.box.com/v/ChestXray-NIHCC][1]\nDid I understand something wrong?\n\n [1]: https://nihcc.app.box.com/v/ChestXray-NIHCC", "387559": "https://github.com/fchollet/deep-learning-models/releases", "382058": "https://data.mendeley.com/datasets/rscbjbr9sj/3", "381789": "https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia", "381510": "https://www.kaggle.com/nih-chest-xrays/data", "380064": "keras pre-trained models\nhttps://keras.io/applications/", "378885": "The question has come up around use of other annotated datasets that are based off of the same set of underlying chest x-ray images from the NIH. First, a clarification that these other public datasets annotated on the NIH images are **not** the labels used in any part of this competition's provided train or test sets. The host team independently labeled the images for this competition specifically. In addition, as per the competition's rules, \"Entrants may re-annotate images in the training set, but may not hand-label predictions, including having human observers rate and evaluate the test data set.\" This means that any hand annotations - including external data which has been annotated as such - are only permitted to be used on the **training set** of images. Solutions that make use of any form of annotation on the original NIH images outside of the training set may be grounds for disqualification. ", "377772": "Are there requirements for posting pre-trained models?", "406333": "", "397423": "https://www.kaggle.com/kmader/pulmonary-chest-xray-abnormalities", "379223": "" } }