How can I contribute?

[jaxony]

I'd love to help out. Are there any ways we can contribute to Open-Reid?

[Cysu]

@jaxony Thank you very much for your interest in the project! I will compile a list of TODOs in detail later this weekend. For now, we may consider some simple changes:

• Support more datasets: PRID, iLIDS, GRID, etc. May refer to this list
• Combining all the datasets together to train models. (This mainly helps for applications)
• More neural network models: ResNeXt, Inception-ResNet-v2, Shake-shake regularization, SqueezeNet, MobileNet, ShuffleNet, etc.
• More metric learning algorithms: XQDA, Null Space Projection, etc.
• Evaluation with multi-crop and multi-scale input images
• Improving the documents

Some more complicated ones include:

• Adding a reranking module with a factory of different methods
• Refactor the feature_extraction module to make it support handcrafted features: LOMO, GOG, etc.
• Support siamese networks with joint feature embedding

Any comments are welcome!

[jaxony]

@Cysu Thanks for the list! It's definitely enough to get started on. I'll start with the simpler changes.

[jaxony]

Hi @Cysu, I implemented a ShuffleNet over the weekend to add to open-reid. Unfortunately I don't have a good GPU to train on ImageNet to see if it even works. I wrote some basic tests, so the model's output size and operations should be correct.

I was hoping maybe you could test it out if you have a good GPU on hand?

[Cysu]

@jaxony That's great! Could you please open a pull request, so that I can checkout locally to test it?