Temporally Distributed Networks for Fast Video Semantic Segmentation

Benchmark Model Rank Results
real-time-semantic-segmentation-on-camvidTD2-PSP50#11mIoU: 76.0Frame (fps): 11(TitanX)Time (ms): 90
real-time-semantic-segmentation-on-camvidTD4-PSP18#18mIoU: 72.6Frame (fps): 25(TitanX)Time (ms): 40
real-time-semantic-segmentation-on-cityscapesTD4-BISE18#15mIoU: 74.9%Frame (fps): 47.6 (Titan X)Time (ms): 21
real-time-semantic-segmentation-on-nyu-depth-1TD2-PSP50#4mIoU: 43.5Speed(ms/f): 35
real-time-semantic-segmentation-on-nyu-depth-1TD4-PSP18#9mIoU: 37.4Speed(ms/f): 19
semantic-segmentation-on-nyu-depth-v2TD2-PSP50#70Mean IoU: 43.5
semantic-segmentation-on-nyu-depth-v2TD4-PSP18#81Mean IoU: 37.4
semantic-segmentation-on-urbanlfTDNet (ResNet-50)#12mIoU (Syn): 74.71mIoU (Real): 76.48
video-semantic-segmentation-on-camvidTDNet-50#3Mean IoU: 76.2
video-semantic-segmentation-on-cityscapes-valTDNet-50 [9]#2mIoU: 79.9