Pytorch implementation of WUTON: a Warping U-net for a Virtual Try-On system using an agnostic person representation along with it's densepose information.
- The first one is the appaerl image cut out of the person image.
- The second one is the product image of the apparel after geometric transformation based on predicted TPS transformation parameters.
- The third one is the person image.
- The fourth one is the person image reconstructed from an agnostic image, given the apparel product image.
- The sixth one is of the same person wearing a differnt apparel altogether which is the fifth image.
- Linux
- Python3, PyTorch
- NVIDIA GPU (8G memory or larger) + CUDA cuDNN
We have used Person Image along with it's apparel image scapred from zalando website. Along with Person-Apparel pair we have used the segmentation information and desnepose information of the person to train our model.