Skip to content

mreso/PortraitSegmentation

Repository files navigation

Implementation of SINet

SINet implementation

Implementation of the paper "SINet: Extreme Lightweight Portrait Segmentation Networks with Spatial Squeeze Modules and Information Blocking Decoder" for portrait segmentation in PyTorch including pretrained weights and training script reaching 95.25 IoU on the EG1800 dataset (trained with same schedule and data as original authors).

Sample results

Usage

ls -l datasets
EG1800
Nukki
--\
   baidu_V1
   baidu_V2

Install requirements

pip install -r requirements.txt


Run training

python train.py [--skip-encoder] [--use-cuda]


# Todo
Fix issue with different input image size than 244

About

Implementation of SINet

Resources

License

MIT, MIT licenses found

Licenses found

MIT
LICENSE
MIT
license.txt

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages