Pytorch implementation of PSPnet for head/face segmentation(dataset is not allowed to provide according to the authors, but you can ask for a private download link)
This is Pytorch version of PSPNet(adapted from link). Support of pytorch 1.4.0.
- Original implementation uses offical resnet weights, however the resent structure has been modified in the latest pytorch code.(FIXED)
- Add dataloader scripts to read color images, convert segmentation image to label image(pixel values range from 1-9) and generate classification label.(FIXED)
- Add data augmentation includes: scale, flip and rotation. (TESTING)
- Add single image testing. (FIXED)