For Human3.6M, please download from the official website and run the preprocessing script from HCMOCO. The folder structure should look like:
- human3.6m
- protocol_1/
- rgb
- seg
- flist_2hz_train.txt
- flist_2hz_eval.txt
For LIP & CIHP & ATR, please download from the official website (Section 2.2 Multi-Person). The train_id.txt and val_id.txt can be found from the parsing folder.
After copying the train_idx.txt,
LIP folder structure should look like:
- LIP
-data
-train_id.txt
-train_images
-1000_1234574.jpg
-val_images
-val_id.txt
-Trainval_parsing_annotations
-train_segmentations
-1000_1234574.png
CIHP folder structure should look like:
- CIHP
-instance-level_human_parsing
-Training
-Images
-0008522.jpg
-Category_ids
-0008522.png
-train_id.txt
-Validation
-val_id.txt
ATR folder structure should look like:
- ATR
-humanparsing
-JPEGImages
-SegmentationClassAug
-train_id.txt
-val_id.txt
For VIP, please download VIP_Fine from the official website (Section 2.3 Video Multi-Person Human Parsing). We only use it for training, and the train_id.txt can be found from the parsing folder. After copying the train_idx.txt, the folder structure should look like:
- VIP
- Images/
- Annotations/
- Category_ids/
- videos1/
- 000000000001.png
- ...
- train_id.txt
For ModaNet, please download from the official website. We only use it for training, and the train_id.txt can be found from the parsing folder. After copying the train_idx.txt, the folder structure should look like:
- ModaNet
- images/
- annotations/
- train_id.txt
For DeepFashion2, please download from the official website for train.zip and validation.zip. We only use it for training, and the train_id.txt can be found from the parsing folder. After copying the train_idx.txt, the folder structure should look like:
- DeepFashion2
- train/
- validation/
- train_id.txt