Scene parsing
End to end semnatic segmantation using deeplab v3 trained on ade20k
Installation
using Docker
docker build -t scene-parsing:latest .
without Docker
virtualenv env
source env/bin/activate
pip install -r requirements.txt
Running
with Docker
docker run -d -p 3030:3030 scene-parsing
without docker
python api.py
Results
mIOU: 45.65% (val)
Pixel-wise Accuracy: 82.52% (val)
File Size: 460 mb
image size: 512 * 512