VisionGNN: Dataset+Model+Unit tests+Example

[nam-hoang-le]

🚀 The feature, motivation and pitch

Paper: https://arxiv.org/abs/2206.00272
Motivation: I saw it when doing my own project. From the results in the original paper, we can see its potential to replace CNN-based and Transformer-based models in Vision task due to its efficiency and good performance. Otherwise, we can apply a lot of Graph theories into the field.
Goal: Try to reproduce the work done in this paper while typing it as closely to the existing frameworks in PyG.
Note:

• From the MoleculeGPT issue, we should use as many as many existing features as possible from PyG. - Additional features which you feel reusable for other workflows should be added to PyG.
• One-off functions that are specific to this workflow can be left inside the example.

Alternatives

No response

Additional context

No response

[nam-hoang-le]

@puririshi98 hello sir, I just uploaded the issue related to our discussion

[nam-hoang-le]

Here is the short list of what to do:
Dataset:

• ImageNet
• COCO
  Model:
• Need to code out, refer to the original code: https://github.com/huawei-noah/Efficient-AI-Backbones/tree/master/vig_pytorch