VENHEADs / segmentation_models

Segmentation models with pretrained backbones. Keras.

Segmentation models Zoo

Segmentation models with pretrained backbones

Avaliable models:

• Unet
• FPN
• Linknet
• PSPNet

Avaliable backbones:

Backbone model	Name	Weights	Preprocessing
VGG16	vgg16	imagenet	KA*
VGG19	vgg19	imagenet	KA*
ResNet18	resnet18	imagenet	BGR, [0, 255]
ResNet34	resnet34	imagenet	BGR, [0, 255]
ResNet50	resnet50	imagenet imagenet11k-places365ch	BGR, [0, 255]
ResNet101	resnet101	imagenet	BGR, [0, 255]
ResNet152	resnet152	imagenet imagenet11k	BGR, [0, 255]
ResNeXt50	resnext50	imagenet	[0, 255]
ResNeXt101	resnext101	imagenet	[0, 255]
DenseNet121	densenet121	imagenet	KA*
DenseNet169	densenet169	imagenet	KA*
DenseNet201	densenet201	imagenet	KA*
Inception V3	inceptionv3	imagenet	KA*
Inception ResNet V2	inceptionresnetv2	imagenet	KA*

• KA - preprocessing function from keras.applications

Requirements

1. Python 3.6 or higher
2. Keras >=2.1.0
3. Tensorflow >= 1.4

Installation

Installing via pip

$ pip install segmentation_models

Using latest version in your project

$ git clone https://github.com/qubvel/segmentation_models.git
$ cd segmentation_models
$ git submodule update --init --recursive

Code examples

Train Unet model:

from segmentation_models import Unet

# prepare data
x, y = ...

# prepare model
model = Unet(backbone_name='resnet34', encoder_weights='imagenet')
model.compile('Adam', 'binary_crossentropy', ['binary_accuracy'])

# train model
model.fit(x, y)

Train FPN model:

from segmentation_models import FPN

model = FPN(backbone_name='resnet34', encoder_weights='imagenet')

Useful trick

Freeze encoder weights for fine-tuning during first epochs of training:

from segmentation_models import FPN
from segmentation_models.utils import set_trainable

model = FPN(backbone_name='resnet34', encoder_weights='imagenet', freeze_encoder=True)
model.compile('Adam', 'binary_crossentropy', ['binary_accuracy'])

# pretrain model decoder
model.fit(x, y, epochs=2)

# release all layers for training
set_trainable(model) # set all layers trainable and recompile model

# continue training
model.fit(x, y, epochs=100)

TODO

• Update Unet API
• Update FPN API
• Add Linknet models
• Add PSP models
• Add DPN backbones

Change Log

Version 0.1.1

• Added Linknet model
• Keras 2.2+ compatibility (fixed import of _obtain_input_shape)
• Small code improvements and bug fixes

Version 0.1.0

• Unet and FPN models