Home

Awesome

ECCV2018 - Learning Human-Object Interactions by Graph Parsing Neural Networks

Introduction

The algorithm is described in the ECCV 2018 paper: Learning Human-Object Interactions by Graph Parsing Neural Networks. In this work, we introduce the Graph Parsing Neural Network (GPNN), a framework that incorporates structural knowledge while being differentiable end-to-end. teaser


Environment and installation

This repository is developed under CUDA8.0 and pytorch3.1 in python2.7. Early versions of pytorch can be found here. The required python packages can be installed by:

pip install http://download.pytorch.org/whl/cu80/torch-0.3.1-cp27-cp27mu-linux_x86_64.whl
pip install -r requirements.txt

Download the pre-trained model and features

/gpnn
  /src
    /python
  /tmp
    /cad120
    /hico
      /processed
        /hico_data_background_49
    /checkpoints
    /results

Running the code

If you want to train the model from scratch, just change the default epoch number from 0 to 100, and rename the pre-trained models. Running the above files should start the training.

Evaluation

The experiment results are provided in tmp/results/. The benchmarking tool for the DET-HICO datset can be found here.


Citation

If you find this code useful, please cite our work with the following bibtex:

@inproceedings{qi2018learning,
    title={Learning Human-Object Interactions by Graph Parsing Neural Networks},
    author={Qi, Siyuan and Wang, Wenguan and Jia, Baoxiong and Shen, Jianbing and Zhu, Song-Chun},
    booktitle={European Conference on Computer Vision (ECCV)},
    year={2018}
}