Another Implementation on CP-VITON-PLUS dataset, extended for 3D reconstruction from single 2D image Dataset
Steps:
- First prepare the dataset
- Train GMM 2.a Requirements: cloth, cloth mask, image, image mask, image parse(segmentation), pose(keypoint detection) 2.b Output: Warped cloth, warped grid, warped mask, overlayed TPS
- Test GMM with testing set
- Train TOM 4.a Requirement: Warped cloth,warped mask, Image, pose, image mask. 4.b Output: Composite mask, Rendered image, Final improved image
- Test TOM
Pitfalls:
- Run test.py for GMM network with the training dataset (not the testing dataset)
- Copy the result of GMM module: "warp-cloth and "warp-mask" folders into "data/train" directory. This will be used to train the TOM as mentioned in Step 4.a above
- For testing your custom images, resize them to 192x256 pixels, and obtain image-parse, cloth mask and pose(keypoints)
Credit for 2d goes to: