ziwei-zh / CorrPM

Pytorch implementation of CVPR2020 paper "Correlating Edge, Pose with Parsing"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Correlating Edge, Pose with Parsing

This is a Pytorch implementation of our paper Correlating Edge, Pose with Parsing accepted by CVPR2020. We propose a Correlation Parsing Machine (CorrPM) utilizing a Heterogeneous Non-Local (HNL) network to capture the correlations among features from human edge, pose and parsing.

Requirements

Pytorch 0.4.1

Python 3.6

The compile of InPlace-ABN is based on CE2P.

Implementation

Dataset

Please download LIP dataset and make them follow this structure:

|-- LIP
    |-- TrainVal_pose_annotations/LIP_SP_TRAIN_annotations.json
    |-- images_labels
        |-- train_images
        |-- train_segmentations
        |-- val_images
        |-- val_segmentations
        |-- train_id.txt
        |-- val_id.txt

Pose annotation file can be downloaded here Google drive.

Train and Test

The pre-trained ResNet-101 can be downloaded from Google drive or Baidu drive.

./run_train.sh

Please download our trained model from Google drive or Baidu drive and put it in the ./pth folder.

./run_eval.sh

Citation

If you use this code for your research, please cite our papers.

@InProceedings{Zhang_2020_CVPR,
author = {Zhang, Ziwei and Su, Chi and Zheng, Liang and Xie, Xiaodong},
title = {Correlating Edge, Pose With Parsing},
booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}
}

About

Pytorch implementation of CVPR2020 paper "Correlating Edge, Pose with Parsing"

License:MIT License


Languages

Language:Python 81.3%Language:Cuda 9.0%Language:C++ 6.1%Language:C 1.9%Language:Shell 1.6%