Keras implementation of Deeplabv2 with tensorflow backend

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.

How to use

Execute python testing.py (Input image is defined in the testing.py, so edit it to use different image).

Requirements (Tested on those versions)

Python==2.7.12 Keras==2.2.4
tensorflow==1.9.0
CUDA==9.0.176

Improvements (TODO)

• add fully-connected CRF post processing (pydensecrf?)
• add ResNet-101 encoder

Segmentation results example