ImplicitSeg

A Pytorch Segmentation module through implicit way (support 2d and 3d)

Install

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

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

Note to run demo.py with --vis option, you also need to additional dependence: vtkplotter and trimesh and skimage.

Prepare Test data

First mkdir ./data, then download image.png(here) and sdf.pth(here), and put them under ./data.

Usage

# 2d
python test/check_seg2d.py --mask ./data/image.png --mode lossless --vis
python test/check_seg2d.py --mask ./data/image.png --mode topk --vis
# 3d
python test/check_seg3d.py --voxel ./data/sdf.pth --mode lossless --vis
python test/check_seg3d.py --voxel ./data/sdf.pth --mode topk --vis

Note: Seg3dTopk and Seg3dLossless are both instances of nn.Module, so you need to be carefull when you want to integrate this to other trainable model.