[AAAI 2024] Large Occluded Human Image Completion via Image-Prior Cooperating

Requirements

pytorch
torchvision
numpy
random
glob
einops
timm
pyiqa
opencv

Prepare dataset

Download AHP dataset https://sydney0zq.github.io/ahp/
Download object mask https://maildluteducn-my.sharepoint.com/:u:/g/personal/zengyu_mail_dlut_edu_cn/EUcTzYoX5LhFkBCjG62E5wwBzru1eM4PhwmRNGmo08pH5Q?download=1
Run python crop_AHP_dataset.py to crop the image
Run python dataset/crop_obj_mask.py to crop the object mask for test
Run python dataset/generate_mask_for_test.py to generate the center mask for test
Move all images with the word test in the file name to dataset/test/

Train

1. Pre-training python train_pretrain.py
2. Train the coarse network python train_corse.py
3. Train the refinement network python train_refinement.py
4. Train the segmentation network python train_unet_seg.py

Test

1. We provided a trained weight:https://drive.google.com/file/d/1cns3WYNf0lKQytzDOi2BlQF8P7WdhCsb/view?usp=sharing and a trained UNet:https://drive.google.com/file/d/1TQ5jmvqoMVPtZ42Dlxc6UkOiLxCUhOFm/view?usp=sharing
2. Test python test.py