Temporally Distributed Networks for Fast Video Semantic Segmentation

3 Apr 2020Ping HuFabian Caba HeilbronOliver WangZhe LinStan SclaroffFederico Perazzi

We present TDNet, a temporally distributed network designed for fast and accurate video semantic segmentation. We observe that features extracted from a certain high-level layer of a deep CNN can be approximated by composing features extracted from several shallower sub-networks... (read more)

PDF Abstract
TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT LEADERBOARD
Real-Time Semantic Segmentation CamVid TD2-PSP50 mIoU 76.0 # 3
Time (ms) 90 # 5
Frame (fps) 11(TitanX) # 8
Real-Time Semantic Segmentation CamVid TD4-PSP18 mIoU 72.6 # 5
Time (ms) 40 # 4
Frame (fps) 25(TitanX) # 8
Real-Time Semantic Segmentation Cityscapes test TD4-BISE18 mIoU 74.9% # 4
Time (ms) 21 # 9
Frame (fps) 47.6 (Titan X) # 17
Real-Time Semantic Segmentation NYU Depth v2 TD4-PSP18 mIoU 37.4 # 4
Speed(ms/f) 19 # 1
Semantic Segmentation NYU Depth v2 TD4-PSP18 Mean IoU 37.4 # 4
Semantic Segmentation NYU Depth v2 TD2-PSP50 Mean IoU 43.5 # 2
Real-Time Semantic Segmentation NYU Depth v2 TD2-PSP50 mIoU 43.5 # 1
Speed(ms/f) 35 # 2