Demos

Light-Weight RefineNet in TensorFlow.js

Description

This demo uses the pretrained Light-Weight RefineNet and Multi-Task RefineNet models converted from PyTorch. The first model is trained on PASCAL VOC to segment 20 classes and background, the second model is trained on NYUD-v2 and is able to segment 40 classes, predict depth in the range of 8 meters and estimate surface normals.

The demo uses the WebGL backend for inference. If nothing happens after uploading an image or the output does not look reasonable at all, it means that your device is not yet supported. Please see console message for more information.

Input / Model Outputs

Upload an image:

Segmentation Colour Map

Input / Model Outputs

Upload an image:

Segmentation Colour Maps