Human Instance Segmentation

A Single-Human Instance Segmentor runs at 50 FPS on GV100. Both training and inference is included in this repo.

Install

# via pip
pip install git+https://github.com/Project-Splinter/human_inst_seg --upgrade

# via git clone
git clone https://github.com/Project-Splinter/human_inst_seg
cd human_inst_seg
python setup.py develop

Note to run demo.py, you also need to install streamer_pytorch through:

pip install git+https://github.com/Project-Splinter/streamer_pytorch --upgrade

Train

First Download dataset from here

git clone https://github.com/Project-Splinter/human_inst_seg; cd human_inst_seg;
mkdir ./data # put all dataset zip under here and unzip them. It should contain two folders: `ATR_RemoveBG` and `alignment`
python human_inst_seg/train.py

Usage

# images
python demo.py --images <IMAGE_PATH> <IMAGE_PATH> <IMAGE_PATH> --loop --vis
# videos
python demo.py --videos <VIDEO_PATH> <VIDEO_PATH> <VIDEO_PATH> --vis
# capture device
python demo.py --camera --vis

API

seg_engine = Segmentation(ckpt=None, device="cuda:0", init=True):
seg_engine.init(pretrained="")
seg_engine.forward(input)  

Note: Segmentation is an instance of nn.Module, so you need to be carefull when you want to integrate this to other trainable model.