I was looking to create a cycle gan or implement this paper using Fastai m2e-try on . Read through a lot of stuff (https://hardikbansal.github.io/CycleGANBlog/) , (https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix) and the recent Caffe based Facebookreasearch gihutb [detector2 , densepose].
Looking forward for some leads to implement this paper ,
- Extract pose from the input image
- Extract pose from the output image
- Convert the pose of output image to input.
- Extract fabric from input image
- Align the fabric to the output image.
The example of horse2zerbra did not work well as the input/output image fabric will keep changing.
Incase I have choosen a wrong category to post please correct me.