{ "id": 373211, "title": "VinDr-Mammo: 337.8 GB 5000 patients with full-field digital mammography and yolov5 models trained on it", "url": "/competitions/rsna-breast-cancer-detection/discussion/373211", "author_name": "@kaggleqrdl", "post_date": "2022-12-20T07:59:05.123000", "votes": 33, "comment_count": 16, "views": 0, "content": "

This has been quietly mentioned in a couple of places, but I'd like to highlight it more clearly.

\n

Link here - https://physionet.org/content/vindr-mammo/1.0.0/

\n

Features:

\n\n

For simple malignant tumour bbox training, likely only the 241 findings marked BIRADS 5 can be used.

\n

However, a large number of papers I've read follow a two staged approach, where the first stage is segmenting / finding ROI, and the second stage (if there is one) is evaluating the region(s) for malignancy.

\n

So I do believe the other 2000 region marked images in Vindr are useful for training on yolo models and/or other approaches. @awsaf49 also has some ideas around collecting ROI/bbox data via gradcam which can be used in combination with this.

\n

Note, I don't think I've run across very many papers which are evaluating on whole breasts having cancer in the way this competition is. That said, I welcome and encourage falsification, at least via cites published in high impact factor journals :)

\n

For the yolov5 info, check out my posts elsewhere. Apparently I've been spamming =)

", "messages": [ { "id": 2070613, "postDate": "2022-12-20T07:59:05.123Z", "content": "

This has been quietly mentioned in a couple of places, but I'd like to highlight it more clearly.

\n

Link here - https://physionet.org/content/vindr-mammo/1.0.0/

\n

Features:

\n\n

For simple malignant tumour bbox training, likely only the 241 findings marked BIRADS 5 can be used.

\n

However, a large number of papers I've read follow a two staged approach, where the first stage is segmenting / finding ROI, and the second stage (if there is one) is evaluating the region(s) for malignancy.

\n

So I do believe the other 2000 region marked images in Vindr are useful for training on yolo models and/or other approaches. @awsaf49 also has some ideas around collecting ROI/bbox data via gradcam which can be used in combination with this.

\n

Note, I don't think I've run across very many papers which are evaluating on whole breasts having cancer in the way this competition is. That said, I welcome and encourage falsification, at least via cites published in high impact factor journals :)

\n

For the yolov5 info, check out my posts elsewhere. Apparently I've been spamming =)

", "rawMarkdown": "This has been quietly mentioned in a couple of places, but I'd like to highlight it more clearly.\n\nLink here - https://physionet.org/content/vindr-mammo/1.0.0/\n\n**Features:**\n- FFDM / Full Field Digital Mammography / similar to RSNA dataset\n- 5000 patients / 337.8 GB dataset\n- All 20K images are marked with density classification\n- Findings annotations with bounding boxes around various types of marked regions. \n - 241 of these findings are marked BIRADS 5 (very high probability of malignancy)\n - 995 of the findings are marked BIRADS 4 (about 30% chance of cancer)\n - Remainder of bbox findings are BIRADS 3, so 2254 images have been annotated\n- I traded emails with who I reasonably believe is the author of the dataset (Nguyễn Quý Hà), and he said it would be OK to use this data with this RSNA Kaggle competition. Still waiting to hear back from physionet. It goes without saying of course, that you can't rely on anything I say here as any type of legal guarantee.\n\nFor simple malignant tumour bbox training, likely only the 241 findings marked BIRADS 5 can be used.\n\nHowever, a large number of papers I've read follow a two staged approach, where the first stage is segmenting / finding ROI, and the second stage (if there is one) is evaluating the region(s) for malignancy. \n\nSo I do believe the other 2000 region marked images in Vindr are useful for training on yolo models and/or ***other*** approaches. @awsaf49 also has some ideas around collecting ROI/bbox data via gradcam which can be used in combination with this.\n\nNote, I don't think I've run across very many papers which are evaluating on whole breasts having cancer in the way this competition is. That said, I welcome and encourage falsification, at least via cites published in high impact factor journals :)\n\nFor the yolov5 info, check out my posts elsewhere. Apparently I've been spamming =)", "votes": 33 }, { "id": 2112882, "postDate": "2023-01-23T22:54:47.193Z", "content": "

As the competition sponsors, we wanted to respond to this thread, somewhat belatedly, to remind competitors that use of public datasets for training is allowed, so long as such datasets are “publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants” (Rules, Section 7.c). The question has been raised whether this rule allows use of datasets such as the VinDR-Mammo dataset that require users to accept certain restrictions on use. The position of the sponsors is that datasets should be considered “publicly available and equally accessible” so long as their restrictions are not substantially greater than those imposed for use of the challenge dataset. In our reading, the restrictions imposed by the license for the VinDR-Mammo dataset are substantially similar to those used for the challenge dataset.

", "rawMarkdown": "As the competition sponsors, we wanted to respond to this thread, somewhat belatedly, to remind competitors that use of public datasets for training is allowed, so long as such datasets are “publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants” (Rules, Section 7.c). The question has been raised whether this rule allows use of datasets such as the VinDR-Mammo dataset that require users to accept certain restrictions on use. The position of the sponsors is that datasets should be considered “publicly available and equally accessible” so long as their restrictions are not substantially greater than those imposed for use of the challenge dataset. In our reading, the restrictions imposed by the license for the [VinDR-Mammo dataset](https://physionet.org/content/vindr-mammo/view-license/1.0.0/) are substantially similar to those used for the challenge dataset.", "votes": 9, "replies": [ { "id": 2112901, "postDate": "2023-01-23T23:34:17.373Z", "content": "

Wow that's great, thanks @cdcarr

", "rawMarkdown": "Wow that's great, thanks @cdcarr ", "votes": 2 }, { "id": 2113018, "postDate": "2023-01-24T03:02:42.103Z", "content": "

I am glad to hear the competition host make this clear. We have a lot of data to deal with. I would like to work on how to handle it effectively in the remaining period. Thank you!

", "rawMarkdown": "I am glad to hear the competition host make this clear. We have a lot of data to deal with. I would like to work on how to handle it effectively in the remaining period. Thank you!", "votes": 2 } ] }, { "id": 2070682, "postDate": "2022-12-20T09:02:34.840Z", "content": "

Sounds like you put a lot of work into this, well done. If this turns out to be usable it could really change this competition.

", "rawMarkdown": "Sounds like you put a lot of work into this, well done. If this turns out to be usable it could really change this competition.", "votes": 3 }, { "id": 2070821, "postDate": "2022-12-20T11:33:29.980Z", "content": "

I also checked this dataset, however I'm not sure if we can use it, can we get answer from the hosts?

", "rawMarkdown": "I also checked this dataset, however I'm not sure if we can use it, can we get answer from the hosts?", "votes": 2, "replies": [ { "id": 2070936, "postDate": "2022-12-20T13:34:28.047Z", "content": "

I agree very much.
\nWe would like a response from the host for confirmation.

\n

@sohier
\nThank you for all your efforts for this competition.
\nIs this dataset available?
\nIt would be great for us to get an answer. Thanks!

", "rawMarkdown": "I agree very much.\nWe would like a response from the host for confirmation.\n\n\n@sohier \nThank you for all your efforts for this competition.\nIs this dataset available?\nIt would be great for us to get an answer. Thanks!", "votes": 2 }, { "id": 2071194, "postDate": "2022-12-20T18:23:14.920Z", "content": "

Yes, it would be great to know because in the data use agreement here:
\nhttps://physionet.org/content/vindr-mammo/view-dua/1.0.0/
\nWe can see:

\n
\n

I will not share access to PhysioNet restricted data with anyone else.

\n
", "rawMarkdown": "Yes, it would be great to know because in the data use agreement here:\nhttps://physionet.org/content/vindr-mammo/view-dua/1.0.0/\nWe can see:\n>I will not share access to PhysioNet restricted data with anyone else.", "votes": 1, "replies": [ { "id": 2071318, "postDate": "2022-12-20T21:39:03.797Z", "content": "
\n

Freely & publicly available external data is allowed, including pre-trained models.

\n
\n

So it must be available for everybody …. not for one person …

", "rawMarkdown": ">Freely & publicly available external data is allowed, including pre-trained models.\n\nSo it must be available for everybody .... not for one person ... ", "votes": 1 }, { "id": 2071327, "postDate": "2022-12-20T21:57:01.743Z", "content": "

Obviously I'm neither the owner of this data nor Kaggle staff, but I don't see why it can't be used.

\n

When Nguyễn Quý Hà replied to my query regarding this, he said \"Yes, the competitors are welcome to use our VinDr-Mammo dataset for this competition.\".

\n

His goals, RSNA goals, and physionet goals are all very aligned afaict - openly solve the breast cancer detection problem and save lives through preventative scanning.

\n

I suppose RSNA might want there to be an option for private corporate use of the data, but that would be rather tragic, imho. Private corps can still use our code, they'd just have to buy their own data if they want to do their own training.

\n

I encourage @sohier to follow up separately with Nguyễn, and probably just him so as not to overload his inbox.

\n

At the very least though, folks like @vbookshelf and myself can make our models openly available I'm sure. But I'll confirm that if for some reason it turns out other folks can't use the data.

", "rawMarkdown": "Obviously I'm neither the owner of this data nor Kaggle staff, but I don't see why it can't be used. \n\nWhen Nguyễn Quý Hà replied to my query regarding this, he said \"Yes, the competitors are welcome to use our VinDr-Mammo dataset for this competition.\". \n\nHis goals, RSNA goals, and physionet goals are all very aligned afaict - openly solve the breast cancer detection problem and save lives through preventative scanning. \n\nI suppose RSNA might want there to be an option for private corporate use of the data, but that would be rather tragic, imho. Private corps can still use our code, they'd just have to buy their own data if they want to do their own training.\n\nI encourage @sohier to follow up separately with Nguyễn, and probably just him so as not to overload his inbox. \n\nAt the very least though, folks like @vbookshelf and myself can make our models openly available I'm sure. But I'll confirm that if for some reason it turns out other folks can't use the data. " }, { "id": 2071928, "postDate": "2022-12-21T14:23:15.073Z", "content": "

I think the main issue is that the PhysioNet DUA may conflict with the rules of the competition:

\n

PhysioNet DUA (https://www.physionet.org/content/vindr-mammo/view-dua/1.0.0/)

\n
\n

I will use the data for the sole purpose of lawful use in scientific research and no other.

\n
\n

Competition Rules:

\n
\n
    \n
  1. WINNER LICENSE.
    \n Open Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code.
    2. \n
\n
\n

If we use a model trained on the VinDr-Mammo dataset, does that now limit commercial use of the model? What if we indirectly use that model to develop our final solution, but the model itself is not part of it?

", "rawMarkdown": "I think the main issue is that the PhysioNet DUA may conflict with the rules of the competition:\n\nPhysioNet DUA (https://www.physionet.org/content/vindr-mammo/view-dua/1.0.0/)\n>I will use the data for the **sole purpose of lawful use in scientific research** and no other.\n\nCompetition Rules:\n>1. WINNER LICENSE.\nOpen Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Open Source Initiative-approved license (see www.opensource.org) that **in no event limits commercial use of such code or model containing or depending on such code.**\n\nIf we use a model trained on the VinDr-Mammo dataset, does that now limit commercial use of the model? What if we indirectly use that model to develop our final solution, but the model itself is not part of it? ", "votes": 2 }, { "id": 2072006, "postDate": "2022-12-21T15:42:35.547Z", "content": "

The prizes page says making the model available is optional, though strongly suggested.

\n

It would be an extremely sad state for everyone involved if folks get hung up on that point. We could potentially end up with sub optimal models which end up less relevant because of bureaucratic silliness.

\n

And, to be frank, I don't think our trained models will have private applications beyond research. The performance constraints in this contest are quirky and likely what will be of interest will be how we technically get around them to develop sufficiently interesting solutions.

\n

But I admit I'm probably more hopeful than optimistic this will get settled reasonably. Happy to see the vindr folks don't have such qualms tho.

", "rawMarkdown": "The prizes page says making the model available is optional, though strongly suggested. \n\nIt would be an extremely sad state for everyone involved if folks get hung up on that point. We could potentially end up with sub optimal models which end up less relevant because of bureaucratic silliness. \n\nAnd, to be frank, I don't think our trained models will have private applications beyond research. The performance constraints in this contest are quirky and likely what will be of interest will be how we technically get around them to develop sufficiently interesting solutions.\n\nBut I admit I'm probably more hopeful than optimistic this will get settled reasonably. Happy to see the vindr folks don't have such qualms tho.\n\n", "votes": 2 } ] } ] }, { "id": 2183728, "postDate": "2023-03-15T21:31:14.483Z", "content": "

where i can download vindr:mammo from?

", "rawMarkdown": "where i can download vindr:mammo from?" }, { "id": 2091003, "postDate": "2023-01-07T22:28:19.320Z", "content": "

Looks like vindr zip has been uploaded to Kaggle https://www.kaggle.com/datasets/tommyngx/vindrmammo1

", "rawMarkdown": "Looks like vindr zip has been uploaded to Kaggle https://www.kaggle.com/datasets/tommyngx/vindrmammo1", "replies": [ { "id": 2093176, "postDate": "2023-01-09T21:02:38.040Z", "content": "

fyi - https://www.kaggle.com/discussions/product-feedback/375132#2093168

\n
\n

Using your workaround, please continue with uploading the dataset privately. Our support team has reviewed the license and the public version that you found will be taken down. If the workaround of renaming the file extension will suffice, maybe that's the easiest way for now.

\n
\n

Sad state of affairs, tbh, though hardly kaggles fault. Physionet folks need a more permissive license for comps.

", "rawMarkdown": "fyi - https://www.kaggle.com/discussions/product-feedback/375132#2093168\n\n>Using your workaround, please continue with uploading the dataset privately. Our support team has reviewed the license and the public version that you found will be taken down. If the workaround of renaming the file extension will suffice, maybe that's the easiest way for now.\n\nSad state of affairs, tbh, though hardly kaggles fault. Physionet folks need a more permissive license for comps.", "votes": 1 } ] }, { "id": 2071373, "postDate": "2022-12-21T00:40:02.160Z", "content": "

If this turns out to be usable, people with poor GPU machine like me could give up this competition.😂

", "rawMarkdown": "If this turns out to be usable, people with poor GPU machine like me could give up this competition.😂", "replies": [ { "id": 2071392, "postDate": "2022-12-21T01:20:01.883Z", "content": "

@fire15, you can get access to https://vast.ai/#pricing .. 4x RTX A6000 80c / hour. I've used it before, it seems pretty good. You need to be somewhat careful about only exposing as little code as possible as it's not super secure. Using wandb can help with training across interrupted / multiple instances, just remember to revoke your token.

\n

I'll be donating some training for free as well that everyone can use. Hopefully folks will contribute to this effort in terms of ideas.

", "rawMarkdown": "@fire15, you can get access to https://vast.ai/#pricing .. 4x RTX A6000 80c / hour. I've used it before, it seems pretty good. You need to be somewhat careful about only exposing as little code as possible as it's not super secure. Using wandb can help with training across interrupted / multiple instances, just remember to revoke your token.\n\nI'll be donating some training for free as well that everyone can use. Hopefully folks will contribute to this effort in terms of ideas. ", "votes": 2 } ] } ], "comments": [ { "id": 2112882, "author_name": "Chris Carr", "author_url": "", "post_date": "2023-01-23T22:54:47.193000", "content": "

As the competition sponsors, we wanted to respond to this thread, somewhat belatedly, to remind competitors that use of public datasets for training is allowed, so long as such datasets are “publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants” (Rules, Section 7.c). The question has been raised whether this rule allows use of datasets such as the VinDR-Mammo dataset that require users to accept certain restrictions on use. The position of the sponsors is that datasets should be considered “publicly available and equally accessible” so long as their restrictions are not substantially greater than those imposed for use of the challenge dataset. In our reading, the restrictions imposed by the license for the VinDR-Mammo dataset are substantially similar to those used for the challenge dataset.

", "votes": 9, "replies": [ { "id": 2112901, "author_name": "@kaggleqrdl", "author_url": "", "post_date": "2023-01-23T23:34:17.373000", "content": "

Wow that's great, thanks @cdcarr

", "votes": 2, "replies": [] }, { "id": 2113018, "author_name": "YYama", "author_url": "", "post_date": "2023-01-24T03:02:42.103000", "content": "

I am glad to hear the competition host make this clear. We have a lot of data to deal with. I would like to work on how to handle it effectively in the remaining period. Thank you!

", "votes": 2, "replies": [] } ] }, { "id": 2070682, "author_name": "Ivan Aerlic", "author_url": "", "post_date": "2022-12-20T09:02:34.840000", "content": "

Sounds like you put a lot of work into this, well done. If this turns out to be usable it could really change this competition.

", "votes": 3, "replies": [] }, { "id": 2070821, "author_name": "slime", "author_url": "", "post_date": "2022-12-20T11:33:29.980000", "content": "

I also checked this dataset, however I'm not sure if we can use it, can we get answer from the hosts?

", "votes": 2, "replies": [ { "id": 2070936, "author_name": "YYama", "author_url": "", "post_date": "2022-12-20T13:34:28.047000", "content": "

I agree very much.
\nWe would like a response from the host for confirmation.

\n

@sohier
\nThank you for all your efforts for this competition.
\nIs this dataset available?
\nIt would be great for us to get an answer. Thanks!

", "votes": 2, "replies": [] }, { "id": 2071194, "author_name": "MPWARE", "author_url": "", "post_date": "2022-12-20T18:23:14.920000", "content": "

Yes, it would be great to know because in the data use agreement here:
\nhttps://physionet.org/content/vindr-mammo/view-dua/1.0.0/
\nWe can see:

\n
\n

I will not share access to PhysioNet restricted data with anyone else.

\n
", "votes": 1, "replies": [ { "id": 2071318, "author_name": "Remek Kinas", "author_url": "", "post_date": "2022-12-20T21:39:03.797000", "content": "
\n

Freely & publicly available external data is allowed, including pre-trained models.

\n
\n

So it must be available for everybody …. not for one person …

", "votes": 1, "replies": [] }, { "id": 2071327, "author_name": "@kaggleqrdl", "author_url": "", "post_date": "2022-12-20T21:57:01.743000", "content": "

Obviously I'm neither the owner of this data nor Kaggle staff, but I don't see why it can't be used.

\n

When Nguyễn Quý Hà replied to my query regarding this, he said \"Yes, the competitors are welcome to use our VinDr-Mammo dataset for this competition.\".

\n

His goals, RSNA goals, and physionet goals are all very aligned afaict - openly solve the breast cancer detection problem and save lives through preventative scanning.

\n

I suppose RSNA might want there to be an option for private corporate use of the data, but that would be rather tragic, imho. Private corps can still use our code, they'd just have to buy their own data if they want to do their own training.

\n

I encourage @sohier to follow up separately with Nguyễn, and probably just him so as not to overload his inbox.

\n

At the very least though, folks like @vbookshelf and myself can make our models openly available I'm sure. But I'll confirm that if for some reason it turns out other folks can't use the data.

", "votes": 0, "replies": [] }, { "id": 2071928, "author_name": "Ian Pan", "author_url": "", "post_date": "2022-12-21T14:23:15.073000", "content": "

I think the main issue is that the PhysioNet DUA may conflict with the rules of the competition:

\n

PhysioNet DUA (https://www.physionet.org/content/vindr-mammo/view-dua/1.0.0/)

\n
\n

I will use the data for the sole purpose of lawful use in scientific research and no other.

\n
\n

Competition Rules:

\n
\n
    \n
  1. WINNER LICENSE.
    \n Open Source: You hereby license and will license your winning Submission and the source code used to generate the Submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code.
    2. \n
\n
\n

If we use a model trained on the VinDr-Mammo dataset, does that now limit commercial use of the model? What if we indirectly use that model to develop our final solution, but the model itself is not part of it?

", "votes": 2, "replies": [] }, { "id": 2072006, "author_name": "@kaggleqrdl", "author_url": "", "post_date": "2022-12-21T15:42:35.547000", "content": "

The prizes page says making the model available is optional, though strongly suggested.

\n

It would be an extremely sad state for everyone involved if folks get hung up on that point. We could potentially end up with sub optimal models which end up less relevant because of bureaucratic silliness.

\n

And, to be frank, I don't think our trained models will have private applications beyond research. The performance constraints in this contest are quirky and likely what will be of interest will be how we technically get around them to develop sufficiently interesting solutions.

\n

But I admit I'm probably more hopeful than optimistic this will get settled reasonably. Happy to see the vindr folks don't have such qualms tho.

", "votes": 2, "replies": [] } ] } ] }, { "id": 2183728, "author_name": "Qx Nam", "author_url": "", "post_date": "2023-03-15T21:31:14.483000", "content": "

where i can download vindr:mammo from?

", "votes": 0, "replies": [] }, { "id": 2091003, "author_name": "@kaggleqrdl", "author_url": "", "post_date": "2023-01-07T22:28:19.320000", "content": "

Looks like vindr zip has been uploaded to Kaggle https://www.kaggle.com/datasets/tommyngx/vindrmammo1

", "votes": 0, "replies": [ { "id": 2093176, "author_name": "@kaggleqrdl", "author_url": "", "post_date": "2023-01-09T21:02:38.040000", "content": "

fyi - https://www.kaggle.com/discussions/product-feedback/375132#2093168

\n
\n

Using your workaround, please continue with uploading the dataset privately. Our support team has reviewed the license and the public version that you found will be taken down. If the workaround of renaming the file extension will suffice, maybe that's the easiest way for now.

\n
\n

Sad state of affairs, tbh, though hardly kaggles fault. Physionet folks need a more permissive license for comps.

", "votes": 1, "replies": [] } ] }, { "id": 2071373, "author_name": "Mr.Fire", "author_url": "", "post_date": "2022-12-21T00:40:02.160000", "content": "

If this turns out to be usable, people with poor GPU machine like me could give up this competition.😂

", "votes": 0, "replies": [ { "id": 2071392, "author_name": "@kaggleqrdl", "author_url": "", "post_date": "2022-12-21T01:20:01.883000", "content": "

@fire15, you can get access to https://vast.ai/#pricing .. 4x RTX A6000 80c / hour. I've used it before, it seems pretty good. You need to be somewhat careful about only exposing as little code as possible as it's not super secure. Using wandb can help with training across interrupted / multiple instances, just remember to revoke your token.

\n

I'll be donating some training for free as well that everyone can use. Hopefully folks will contribute to this effort in terms of ideas.

", "votes": 2, "replies": [] } ] } ], "raw_markdown_by_id": { "2070613": "This has been quietly mentioned in a couple of places, but I'd like to highlight it more clearly.\n\nLink here - https://physionet.org/content/vindr-mammo/1.0.0/\n\n**Features:**\n- FFDM / Full Field Digital Mammography / similar to RSNA dataset\n- 5000 patients / 337.8 GB dataset\n- All 20K images are marked with density classification\n- Findings annotations with bounding boxes around various types of marked regions. \n - 241 of these findings are marked BIRADS 5 (very high probability of malignancy)\n - 995 of the findings are marked BIRADS 4 (about 30% chance of cancer)\n - Remainder of bbox findings are BIRADS 3, so 2254 images have been annotated\n- I traded emails with who I reasonably believe is the author of the dataset (Nguyễn Quý Hà), and he said it would be OK to use this data with this RSNA Kaggle competition. Still waiting to hear back from physionet. It goes without saying of course, that you can't rely on anything I say here as any type of legal guarantee.\n\nFor simple malignant tumour bbox training, likely only the 241 findings marked BIRADS 5 can be used.\n\nHowever, a large number of papers I've read follow a two staged approach, where the first stage is segmenting / finding ROI, and the second stage (if there is one) is evaluating the region(s) for malignancy. \n\nSo I do believe the other 2000 region marked images in Vindr are useful for training on yolo models and/or ***other*** approaches. @awsaf49 also has some ideas around collecting ROI/bbox data via gradcam which can be used in combination with this.\n\nNote, I don't think I've run across very many papers which are evaluating on whole breasts having cancer in the way this competition is. That said, I welcome and encourage falsification, at least via cites published in high impact factor journals :)\n\nFor the yolov5 info, check out my posts elsewhere. Apparently I've been spamming =)", "2112882": "As the competition sponsors, we wanted to respond to this thread, somewhat belatedly, to remind competitors that use of public datasets for training is allowed, so long as such datasets are “publicly available and equally accessible to use by all participants of the Competition for purposes of the competition at no cost to the other participants” (Rules, Section 7.c). The question has been raised whether this rule allows use of datasets such as the VinDR-Mammo dataset that require users to accept certain restrictions on use. The position of the sponsors is that datasets should be considered “publicly available and equally accessible” so long as their restrictions are not substantially greater than those imposed for use of the challenge dataset. In our reading, the restrictions imposed by the license for the [VinDR-Mammo dataset](https://physionet.org/content/vindr-mammo/view-license/1.0.0/) are substantially similar to those used for the challenge dataset.", "2070682": "Sounds like you put a lot of work into this, well done. If this turns out to be usable it could really change this competition.", "2070821": "I also checked this dataset, however I'm not sure if we can use it, can we get answer from the hosts?", "2183728": "where i can download vindr:mammo from?", "2091003": "Looks like vindr zip has been uploaded to Kaggle https://www.kaggle.com/datasets/tommyngx/vindrmammo1", "2071373": "If this turns out to be usable, people with poor GPU machine like me could give up this competition.😂" } }