This is my repo used for unsupervised part segmentation and keypoints discovery.
It includes the implementation of our paper:
Unsupervised Part Segmentation through Disentangling Appearance and Shape. CVPR 2021
as well as a re-implementation of IMM:
Unsupervised Learning of Object Landmarks through Conditional Image Generation (NeurIPS) 2018
# part segmentation (UnsupPartSeg)
python train_SCOPSP.py \
-s path/to/log/dir \
-c configs/SCOPSP/SCOPSP_CUB_BG_Norm_128x128_TPS_PERCEPNEW_ARC_10lm_hm32_newtc.py \
-g 0 # gpu id(s)
# keypoints discovery (IMM, keypoints discovery)
python train_IMM.py \
-s path/to/log/dir \
-c configs/IMM/MyIMM_WILDCELEBA_128x128_TPS_PERCEP_10lm_h32_mask.py \
-g 2,3 # gpu id(s)