Mednist GAN Bundle

[ericspod]

Fixes # .

Description

This is the bundle for the MedNIST GAN network based off the example notebooks in the tutorials repo. I've followed the code pretty closely with that found in the notebooks with a few changes that don't impact results. There are two scripts for training and inference.

I have as the license the copyright information that is at the top of our files and notebooks but I'm open to changing this to something more appropriate.

I've also included the weights for the network as a pt file directly rather than as a download link because it's so small. I feel we're ok with tiny networks to do this but it can be changed if that's better.

Status

Ready

Please ensure all the checkboxes:

• Codeformat tests passed locally by running ./runtests.sh --codeformat.
• In-line docstrings updated.
• Update version and changelog in metadata.json.
• Ensure versions of packages such as monai, pytorch and numpy are correct in metadata.json.
• Descriptions should be consistent with the content, such as eval_metrics of the provided weights and TorchScript modules.
• Files larger than 25MB are excluded and replaced by providing download links in large_file.yml.

[yiheng-wang-nv]

As for the license part, should MedNIST related license be mentioned as well? Hi @Nic-Ma , may need your help to double confirm it (for other bundles, the dataset related license is added).

[ericspod]

I wasn't sure what the license should be for this one so yes some help on it would be good.

[yiheng-wang-nv]

I wasn't sure what the license should be for this one so yes some help on it would be good.

Hi @ericspod , Nic and I discussed about the license, and could you please refer to the first cell of this notebook: https://colab.research.google.com/drive/1wy8XUSnNWlhDNazFdvGBHLfdkGvOHBKe and use the same license?

[ericspod]

Hi @ericspod , Nic and I discussed about the license, and could you please refer to the first cell of this notebook: https://colab.research.google.com/drive/1wy8XUSnNWlhDNazFdvGBHLfdkGvOHBKe and use the same license?

The content in that cell isn't a license, it's stating that the data is licensed under CC but just points to that license's text. The license that should be here is for the network itself, not the data. I've stated the data's licensing conditions in the README.md file so that should suffice for that, but we don't want to assert that the network is copyrighted by the same author as the data and distributed under their terms. What I put initially in the license file is our copyright header but that isn't sufficient either, so the next best thing I would say is to use the Apache 2.0 license MONAI uses?

[Nic-Ma]

Hi @wyli ,

Could you please help share some comments about the license question?
I am also not sure what licenses we should put for this bundle.
Others look good to me now.

Thanks in advance.

[Nic-Ma]

Hi @ericspod ,

About the license, I think maybe you can refer to the BERT model in the huggingface:
https://huggingface.co/bert-base-uncased/blob/main/LICENSE
We may need some guide or policy for developers to write license?

Thanks.

[ericspod]

I think we do need to consider what our licensing should be and what to recommend to contributors. The BERT link uses Apache 2.0 like MONAI, I'd be fine with that. We could also use MIT license that's more permissive and I'd be fine with that as well, but we should definitely establish a policy.

[Nic-Ma]

Hi @ericspod ,

We discussed the license topic with @pdogra89 and the team last week, and updated the contribution guide for it:
https://github.com/Project-MONAI/model-zoo/blob/dev/CONTRIBUTING.md#license-format

Could you please help update this PR for this last part?

Thanks in advance.

[ericspod]

Yes I'll get on updating this, I had wanted to revisit what to do with licensing.

[ericspod]

I've added a statement into the license regarding the MedNIST data, this goes with the license for the network itself. Does this meet the discussed license criteria and the contribution guidelines?

[yiheng-wang-nv]

/build

[Nic-Ma]

Should we put the model.pt in the large_files to reduce the repo size even it's less than 25MB?
Others look good to me.

Thanks for the update.

[ericspod]

Should we put the model.pt in the large_files to reduce the repo size even it's less than 25MB? Others look good to me.

Thanks for the update.

The file is so tiny I don't think it would make much of a difference, I want to have at least one example of a whole bundle in the repo without having to explain our large file system.