Summary

⚠️ Work in progress ⚠️

A collection of semantic image segmentation models implemented in TensorFlow. Contains data-loaders for the generic and medical benchmark datasets.

Hopefully this project will enable researchers to spend less time scaffolding and more time building.

Datasets & Benchmarks

Generic

• ADE20K Scene Parsing : paper
• Microsoft COCO: Common Objects in Context : paper
• Cityscapes : paper
• PASCAL Visual Object Classes : paper
• SUN RGB-D Scene Understanding Benchmark Suite : paper

Medical

• MICCAI - Brain Tumor Image Segmentation Challenge (BRATS)
• MICCAI - Ischemic Stroke Lesion Segmentation (ISLES)

Networks & Models

Generic

• DeepLab v2 : project : C++ code
• RefineNet : MATLAB code
• I-FCN
• FC-DenseNet : theano, lasagne code
• PixelNet : cafffe code
• FCN : slides
• SegNet : caffe code

Medical

• U-Net

Usage

See ./scipts/

Requirements

• Python 2.7
• TensorFlow 0.12+

Resources

Learn

1. TensorFlow Deep Learning Course Get hands on right away with tensorflow and deep learning.
2. Machine Learning, Andrew Ng Deeper dive into basics, less hands .
3. Stanford CS231n videos I can't overstate how fantastic the notes, and videos are.
4. Deep Learning : Book Helpful reference for filling in gaps.
5. Above papers, starting with Fully Convolutional Networks for Semantic Segmentation and video

Code

• TF-Slim
• TF-Slim : Classification Networks
• imagenet-multiGPU.torch
• pixel-cnn++
• NVIDIA Digits Semantic Segmentaiton Example Medical Imaging Example

Contributing

Please do. PEP-8, google style with 2 space idents 🤦️.