Open taozhuo opened 1 year ago
Seconded!
Also, I would be interested in doing the implementation and submitting the PR. @lucidrains is the architecture used in your implementation exactly the same as the original?
Based on the origin paper and this code, the pytorch version of image backbone is MaxViT, different from the efficientnet.
I would also be interested at this. Would like to help if needed. I think here is an example of such conversion, the only thing need to be sure is that 1) both implementations are identical 2) each components in each of the framework with the same name are having the identical structure.
Any progress on this?
someone can provide a model? thanks very much
Google provides pre-trained SavedModel files.