Home

Awesome

Detecting Human-Object Interaction via Fabricated Compositional Learning

This code is based on PMFNet. Thanks for their excellent work! We only change a few parts based on PMFNet and run the code with a single V100 GPU. You can review the code according to the git status.

This code follows the implementation architecture of roytseng-tw/mask-rcnn.pytorch.

Getting Started

Requirements

Tested under python3.

Assume the project is located at $ROOT.

Compilation

Compile the NMS code:

cd $ROOT/lib 
sh make.sh

Data and Pretrained Model Preparation

Create a data folder under the repo,

cd $ROOT
mkdir data

Training

cd $ROOT
sh script/train_vcoco_fcl.sh

Test

cd $ROOT
sh script/test_vcoco_fcl.sh