Rethinking Atrous Convolution for Semantic Image Segmentation

Benchmark Model Rank Results
2d-semantic-segmentation-on-wildscenesDeepLabv3 (ResNet-50)#4mIoU: 43.37mIoU (Temporal DA): 43.95mIoU (Env DA): 36.12
dichotomous-image-segmentation-on-dis-te1DeeplabV3+#20max F-Measure: 0.601weighted F-measure: 0.506MAE: 0.102
dichotomous-image-segmentation-on-dis-te2DeeplabV3+#20max F-Measure: 0.681weighted F-measure: 0.587MAE: 0.105
dichotomous-image-segmentation-on-dis-te3DeeplabV3+#20max F-Measure: 0.717weighted F-measure: 0.623MAE: 0.102
dichotomous-image-segmentation-on-dis-te4DeeplabV3+#19max F-Measure: 0.715weighted F-measure: 0.621MAE: 0.111
dichotomous-image-segmentation-on-dis-vdDeeplabV3+#23max F-Measure: 0.660weighted F-measure: 0.568MAE: 0.114
semantic-segmentation-on-cityscapesDeepLabv3 (ResNet-101, coarse)#39Mean IoU (class): 81.3%
semantic-segmentation-on-cityscapes-valDeepLabv3 (Dilated-ResNet-101)#57mIoU: 78.5%
semantic-segmentation-on-pascal-voc-2012DeepLabv3-JFT#3Mean IoU: 86.9%
semantic-segmentation-on-pascal-voc-2012-valDeepLabv3-JFT#4mIoU: 82.7%
semantic-segmentation-on-selmaDeepLabV3#3mIoU: 70.7