Custom Operators Zoo

[wkcn]

Hi there, this issue is to summarize some custom operators to be supported. Please feel free to add it if you want any operator : )

• [x] RoIAlign
• [x] Softmax
• [x] Standard Convolution
• [ ] Bilinear/Trilinear Interpolation
• [ ] Stand-Alone Self-Attention in Vision Models
• [ ] CTC Loss
• [ ] Deformable Convolution V2
• [ ] Image wrapAffine based on Homography matrix

[kohillyang]

I suggest several operators for reference:

1. CTC Loss, this OP is supported by mxnet, but before PyTorch 1.0, users must compile by themselves.

2. Deformable Convolution (and also DCNV2), Same as 1, users need compiling, and mxnet does not provide CPU implementation version.

3. Image wrapAffine based on Homography matrix. If this one is implemented, users can implement their image augmentations like image crop, translation, rotating and even scaling based on GPU.