TensorFlow Image Segmentation Models
Features (italic means available soon)
- Build segmentation model with selectable architecture and encoder/decoder backbones
- Build 2D and 3D models
- 2 models architectures
- UNet (skip concatenation)
- LinkNet (skip addition)
- 3 backbones building blocks
- VGG-like (basic convolutions with increasing number of filters each level)
- ResNet-like (add the input to the output of the block)
- DenseNet-like
- Create and use your own architecture or backbone easily
Installation
pip install git+git://github.com/Cyril-Meyer/tism
Usage example in example.py
The interesting features of this library are the 3D backbones and the possibility of creating your own backbone. If you want to make 2D image segmentation, I recommend you the very cool qubvel/segmentation_models library which implement more architectures, and pretrained backbones from keras.applications.