DeepLabv2 is one of state-of-art deep learning models for semantic image segmentation.
Based on the paper DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs.
Note: For now only VGG-16 encoder is implemented.
Note2: Weights are stored in the repository using GitLFS. Therefore - it may take some time to clone this repo.
Code is based on the repository DavideA/deeplabv2-keras, which was implemented using Theano backend.
Execute python testing.py
(Input image is defined in the testing.py, so edit it to use different image).
Python==2.7.12
Keras==2.2.4
tensorflow==1.9.0
CUDA==9.0.176
- add fully-connected CRF post processing (pydensecrf?)
- add ResNet-101 encoder