DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs

Benchmark Model Rank Results
semantic-segmentation-on-cityscapesDeepLab-CRF (ResNet-101)#74Mean IoU (class): 70.4%
semantic-segmentation-on-event-basedDeepLab#4mIoU: 71.05
semantic-segmentation-on-pascal-contextDeepLabV2#47mIoU: 45.7
semantic-segmentation-on-pascal-voc-2012DeepLab-CRF (ResNet-101)#29Mean IoU: 79.7%
semantic-segmentation-on-pascal-voc-2012-valDeepLab-CRF (ResNet-101)#13mIoU: 77.69%
semantic-segmentation-on-selmaDeepLabV2#4mIoU: 68.9