2017-CVPR-Scene Parsing through ADE20K dataset

Reference:

B. Zhou, H. Zhao, X. Puig, S. Fidler, A. Barriuso, A. Torralba. Scene Parsing through ADE20k dataset. Proc. CVPR 2017 [pdf] [homepage]

Main contributions:

1. A new scene parsing dataset (ADE20k). This new dataset contains 150 object and stuff classes. Each image has been labeled into objects, parts of objects, and in some cases even parts of parts.
2. A novel Cascade Segmentation network is proposed for parsing the scene.

Overview of the dataset:

• 20,210 training image, 2,000 validation images and 3,000 testing image

Cascade Segmentation Model:

Experiment Results: