ammieqi / occlusion_net

Occlusion-Net: 2D/3D Occluded Keypoint Localization Using Graph Networks

Occlusion-Net: 2D/3D Occluded Keypoint Localization Using Graph Networks

N Dinesh Reddy, Minh Vo, Srinivasa G. Narasimhan

IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019.

[Project] [Paper] [Supp] [Bibtex ]

Live Demo of the algorithm on a youtube live stream can be found below or [HERE ]:

More Results

^{Result of Occlusion-Net on a live video from youtube}

Installation

Setting up with docker

All the stable releases of docker-ce installed from https://docs.docker.com/install/

Install the nvidia-docker from https://github.com/NVIDIA/nvidia-docker

Setting up the docker

nvidia-docker build -t occlusion_net .

Setting up data

You need to fill the Access Form to get a email regarding the dataset and setup at using the following commands:

git clone https://github.com/dineshreddy91/carfusion_to_coco
cd carfusion_to_coco
virtualenv carfusion2coco -p python3.6
source carfusion2coco/bin/activate
pip install cython numpy
pip install -r requirements.txt
python download_carfusion.py (This file need to be downloaded by requesting, please fill to get access to the data)
sh carfusion_coco_setup.sh
deactivate

The final folder format to train on carfusion data needs to look :

Occlusion-Net
   └─datasets
       └─carfusion
           └─train
               └─car_craig1
                   └───images
                       01_0000.jpg
                       01_0001.jpg
                       ...   
                   └───bb
                      01_0000.txt
                      01_0001.txt
                      ...
                   └───gt
                      01_0000.txt   
                      01_0001.txt
                      ...
           └─test
               └─car_penn1
                   └───images
                       01_0000.jpg
                       01_0001.jpg
                       ...   
                   └───bb
                      01_0000.txt
                      01_0001.txt
                      ...
                   └───gt
                      01_0000.txt   
                      01_0001.txt
                      ...
           └─annotations
               car_keypoints_train.json
               car_keypoints_test.json
               

Running with docker

Training the model on the carfusion dataset

sh train.sh occlusion_net <Path_to_Carfusion_dataset>

Testing on a sample image

Download a pretrained model from [Google Drive]

Results on a sample demo image

sh test.sh occlusion_net demo/demo.jpg

Citation

@inproceedings{onet_cvpr19,
author = {Reddy, N. Dinesh and Vo, Minh and Narasimhan, Srinivasa G.},
title = {Occlusion-Net: 2D/3D Occluded Keypoint Localization Using Graph Networks},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
pages = {7326--7335},
year = {2019}
}