EEN

This is an efficient implementation of EEN (Multi-class Human Body Parsing with Edge-Enhancement Network). The code is based upon this implementation.

Download

Plesae download LIP and CIHP dataset.

Pascal-Person-Part dataset and trained models can be found at baidu drive (the password is 'sf5z') or google drive.

Environments

• Python 3.5

• PyTorch 0.4.1

• cffi

• matplotlib

• numpy

• opencv-python

• scipy

• tqdm

• You need to use InPlace-ABN with CUDA implementation, which must be compiled with the following commands:

cd libs
sh build.sh
python build.py

• The model is trained on NVIDIA TITAN RTX 2080 Ti GPU. It will take up about 16G.

Training

• Please set the dataset dir in file 'run.sh'. The contents of each dataset include:

  ─ train_images

  ─ train_labels

  ─ val_images

  ─ val_labels

  ─ train_id.txt

  ─ val_id.txt

• Please put the pretrained resnet101-imagenet.pth in './dataset/'.

• Run the sh run.sh.

Evaluation

If you want to evaluate the trained models on LIP and CIHP, you can run the sh run_evaluate.sh.