keras-io

/

deeplabv3p-resnet50

like 3

Multiclass semantic segmentation using DeepLabV3+

This repo contains the model and the notebook to this Keras example on Multiclass semantic segmentation using DeepLabV3+.

Full credits to: Soumik Rakshit

The model is trained for demonstrative purposes and does not guarantee the best results in production. For better results, follow & optimize the Keras example as per your need.

Background Information

Semantic segmentation, with the goal to assign semantic labels to every pixel in an image, is an essential computer vision task. In this example, we implement the DeepLabV3+ model for multi-class semantic segmentation, a fully-convolutional architecture that performs well on semantic segmentation benchmarks.

Training Data

The model is trained on a subset (10,000 images) of Crowd Instance-level Human Parsing Dataset. The Crowd Instance-level Human Parsing (CIHP) dataset has 38,280 diverse human images. Each image in CIHP is labeled with pixel-wise annotations for 20 categories, as well as instance-level identification. This dataset can be used for the "human part segmentation" task.

Model

The model uses ResNet50 pretrained on ImageNet as the backbone model.

References:

1. Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
2. Rethinking Atrous Convolution for Semantic Image Segmentation
3. DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs