flowtrack.pytorch

Pytorch implementation of FlowTrack.

Simple Baselines for Human Pose Estimation and Tracking (https://arxiv.org/pdf/1804.06208.pdf)

TO DO:

• [x] Human detection
• [x] Single person pose estimation
• [x] Optical flow estimation
• [x] Box propagation
• [ ] Pose tracking

Requirements

pytorch >= 0.4.0
torchvision
pycocotools
tensorboardX

Installation

cd lib
./make.sh

Disable cudnn for batch_norm:

# PYTORCH=/path/to/pytorch
# for pytorch v0.4.0
sed -i "1194s/torch\.backends\.cudnn\.enabled/False/g" ${PYTORCH}/torch/nn/functional.py
# for pytorch v0.4.1
sed -i "1254s/torch\.backends\.cudnn\.enabled/False/g" ${PYTORCH}/torch/nn/functional.py

Training

Pose Estimation

Download data folder as $ROOT/data.

python ./tools/pose/main.py

The official code is released on Microsoft/human-pose-estimation.pytorch.

Demo

Pose Estimation

#TODO

Detection

Download pretrained detection model into models/detection/. Refer to pytorch-faster-rcnn for more information.

python ./tools/detection/demo.py

Optical Flow Estimation

Download pretrained flownet into models/flownet/. Refer to flownet2-pytorch for more information.

python ./tools/flownet/demo.py --model </path/to/model>

Update

2018.12.05:

• Add Pose Estimation Models
• Deconv DenseNet
• Stacked Hourglass Network
• FPN