Addition of other pooling layers

[Saanidhyavats]

@awni We were going through pooling layers available in mlx framework. I think we can add pooling techniques for 3d data such as Avgpooling3d, Maxpooling3d and also implement AdaptivePooling for 2D data.

[Saanidhyavats]

@awni what do you think about it?

[awni]

3D pooling might be useful.. I'm not so sure how commonly used it is. I haven't seen too many requests for AdaptivePooling. In general for features like this I think it's ok to be a bit reactive (add them if needed) rather than adding them just for the sake of feature completeness.

[Saanidhyavats]

3D pooling might be useful.. I'm not so sure how commonly used it is. I haven't seen too many requests for AdaptivePooling. In general for features like this I think it's ok to be a bit reactive (add them if needed) rather than adding them just for the sake of feature completeness.

I was going through the code in https://github.com/apple/ml-cvnets/blob/main/cvnets/layers/adaptive_pool.py and noticed that it's currently leveraging adaptive pooling from PyTorch. I believe we could enhance the project by incorporating adaptive pooling directly into MLX, allowing us to utilize this functionality seamlessly. Would this be something we could explore further?

[awni]

I would suggest putting an issue in that repo about building an MLX back-end (which would be super cool)! If the project adds an MLX back-end and needs support from any missing features we'd be happy to reconsider what to add.

[cvnad1]

@awni We believe that’s a good idea and will consider your advice.