Video Semantic Segmentation on Cityscapes (val)
83.3mIoUMask2Former
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Mask2FormerBackbone=Swin-B2023.10 | 83.3 | — | — | 1,057 | 107 | 3.26 | — | |
| Mask2FormerBackbone=Swin-L2023.10 | 83.3 | — | — | 1,911.3 | 215 | 2.11 | — | |
| Mask2FormerBackbone=Swin-S2023.10 | 82.6 | — | — | 730.1 | 68.7 | 4.31 | — | |
| SegFormerBackbone=MiT-B52023.10 | 82.4 | — | — | 1,460.4 | 84.7 | 7.2 | — | |
| Mask2FormerBackbone=Swin-T2023.10 | 82.1 | — | — | 543.6 | 47.4 | 5.41 | — | |
| MPVSSBackbone=Swin-B2023.10 | 81.7 | — | — | 278.6 | 147 | 9.54 | — | |
| MPVSSBackbone=Swin-L2023.10 | 81.6 | — | — | 449.5 | 255.4 | 7.24 | — | |
| Mask2Former-AccelBackbone=Swin-B+Swin-T2023.10 | 81.4 | — | — | 680.1 | 193.1 | 4.4 | — | |
| MPVSSBackbone=Swin-S2023.10 | 81.3 | — | — | 213.2 | 108 | 10.98 | — | |
| MPVSSBackbone=Swin-T2023.10 | 80.7 | — | — | 175.9 | 114 | 12.33 | — | |
| NetwarpSpeed (fps)=0.33, Avg RT (ms)=3004, MaxLatency (ms)=30042020.07 | 80.6 | — | — | — | — | — | — | |
| TMANet-50 (Ours-50)GFLOPs=7542021.02 | 80.3 | — | — | — | — | — | — | |
| baselineDescription=per-frame segmentation, Backbone=ResNet-1012018.04 | 80.2 | 360 | 360 | — | — | — | — | |
| Mask2FormerBackbone=R1012023.10 | 80.1 | — | — | 685.5 | 63 | 5.68 | — | |
| TD2-PSP50Speed (ms/f)=178, Max Latency (ms)=1782020.04 | 79.9 | — | — | — | — | — | — | |
| TDNet-50GFLOPs=10822021.02 | 79.9 | — | — | — | — | — | — | |
| Mask2Former-DFFBackbone=Swin-B2023.10 | 79.9 | — | — | 525.3 | 145.7 | 6.09 | — | |
| PSPNet101Speed (ms/f)=360, Max Latency (ms)=3602020.04 | 79.7 | — | — | — | — | — | — | |
| Mask2FormerBackbone=R502023.10 | 79.4 | — | — | 529.9 | 44 | 6.58 | — | |
| Mask2Former-AccelBackbone=R101+R502023.10 | 78.9 | — | — | 594.8 | 145.7 | 5.78 | — | |
| PSPNetBackbone=R1012023.10 | 78.5 | — | — | 2,048.9 | 67.9 | 2.88 | — | |
| SegFormerBackbone=MiT-B12023.10 | 78.5 | — | — | 243.7 | 13.8 | 20.7 | — | |
| MPVSSBackbone=R502023.10 | 78.4 | — | — | 173.2 | 84.1 | 13.43 | — | |
| RS-SSMBackbone=MiT-B1, Input size=512 x 10242026.03 | 78.3 | — | — | 59.5 | 20.1 | 27.2 | — | |
| MPVSSBackbone=R1012023.10 | 78.2 | — | — | 204.3 | 103.1 | 12.55 | — | |
| PSPNet50Speed (ms/f)=238, Max Latency (ms)=2382020.04 | 78.1 | — | — | — | — | — | — | |
| Mask2Former-DFFBackbone=R1012023.10 | 77.1 | — | — | 457.4 | 101.7 | 7.14 | — | |
| AFP + AKSFeature Propagation Method=Adaptive (AFP), Key-frame Selection Strategy=Adaptive (AKS), Scheduling Scheme=standard, Backbone=ResNet-1012018.04 | 76.84 | 171 | 380 | — | — | — | — | |
| LVSSpeed (ms/f)=171, Max Latency (ms)=3802020.04 | 76.8 | — | — | — | — | — | — | |
| TD4-PSP18Speed (ms/f)=85, Max Latency (ms)=852020.04 | 76.8 | — | — | — | — | — | — | |
| LVS2021.02 | 76.8 | — | — | — | — | — | — | |
| FANet34+TempSpeed (fps)=58, Avg RT (ms)=17, MaxLatency (ms)=17, Temporal Attention=t=22020.07 | 76.7 | — | — | — | — | — | — | |
| FCNBackbone=R1012023.10 | 76.6 | — | — | 2,203.3 | 68.5 | 2.83 | — | |
| PEARLSpeed (ms/f)=800, Max Latency (ms)=8002020.04 | 76.5 | — | — | — | — | — | — | |
| BiseNet* 101Speed (ms/f)=722020.04 | 76.5 | — | — | — | — | — | — | |
| BiseNet* 101Speed (ms/f)=722020.04 | 76.5 | — | — | — | — | — | — | |
| TD2-Bise34Speed (ms/f)=262020.04 | 76.4 | — | — | — | — | — | — | |
| TD2-Bise34Speed (ms/f)=262020.04 | 76.4 | — | — | — | — | — | — | |
| FANet34Speed (fps)=58, Avg RT (ms)=17, MaxLatency (ms)=172020.07 | 76.3 | — | — | — | — | — | — | |
| BiseNet* 34Speed (ms/f)=272020.04 | 76 | — | — | — | — | — | — | |
| BiseNet* 34Speed (ms/f)=272020.04 | 76 | — | — | — | — | — | — | |
| LVS-LLSSpeed (ms/f)=119, Max Latency (ms)=1192020.04 | 75.9 | — | — | — | — | — | — | |
| Low-LatencySpeed (fps)=7.5, Avg RT (ms)=133, MaxLatency (ms)=1332020.07 | 75.9 | — | — | — | — | — | — | |
| AFP + AKS + LLSFeature Propagation Method=Adaptive (AFP), Key-frame Selection Strategy=Adaptive (AKS), Scheduling Scheme=Low-Latency (LLS), Backbone=ResNet-1012018.04 | 75.89 | 119 | 119 | — | — | — | — | |
| CFFM++Backbone=MiT-B1, Input size=512 x 10242026.03 | 75.7 | — | — | 169.8 | 15.9 | 20.4 | — | |
| TV3SBackbone=MiT-B1, Input size=512 x 10242026.03 | 75.6 | — | — | 83.6 | 17.3 | 25.1 | — | |
| PSPNet18Speed (ms/f)=91, Max Latency (ms)=912020.04 | 75.5 | — | — | — | — | — | — | |
| FANet18+TempSpeed (fps)=72, Avg RT (ms)=14, MaxLatency (ms)=14, Temporal Attention=t=22020.07 | 75.5 | — | — | — | — | — | — | |
| SwiftNetSpeed (ms/f)=232020.04 | 75.4 | — | — | — | — | — | — | |
| SwiftNetSpeed (ms/f)=232020.04 | 75.4 | — | — | — | — | — | — | |
| AFP + fix scheduleFeature Propagation Method=Adaptive (AFP), Key-frame Selection Strategy=fix schedule (l=5), Scheduling Scheme=standard, Backbone=ResNet-1012018.04 | 75.26 | 151 | 360 | — | — | — | — | |
| MRCFABackbone=MiT-B1, Params (M)=14.9, FPS (f/s)=21.52022.07 | 75.1 | — | — | — | — | — | — | |
| CFFM-VSSBackbone=MiT-B12023.10 | 75.1 | — | — | 158.7 | 15.4 | 11.71 | — | |
| MRCFABackbone=MiT-B12023.10 | 75.1 | — | — | 145 | 14.9 | 12.97 | — | |
| MRCFABackbone=MiT-B1, Input size=512 x 10242026.03 | 75.1 | — | — | 145 | 14.9 | 13 | — | |
| TD4-Bise18Speed (ms/f)=212020.04 | 75 | — | — | — | — | — | — | |
| TD4-Bise18Speed (ms/f)=212020.04 | 75 | — | — | — | — | — | — | |
| FANet18Speed (fps)=72, Avg RT (ms)=14, MaxLatency (ms)=142020.07 | 75 | — | — | — | — | — | — | |
| SegFormerBackbone=MiT-B1, Params (M)=13.8, FPS (f/s)=46.82022.07 | 74.1 | — | — | — | — | — | — | |
| SegFormerBackbone=MiT-B1, Input size=512 x 10242026.03 | 74.1 | — | — | 106.4 | 13.7 | 46.8 | — | |
| CFFM-VSSBackbone=MiT-B02023.10 | 74 | — | — | 80.7 | 4.6 | 15.79 | — | |
| CFFMBackbone=MiT-B0, Input size=512 x 10242026.03 | 74 | — | — | 80.7 | 4.6 | 15.8 | — | |
| BiseNet* 18Speed (ms/f)=202020.04 | 73.8 | — | — | — | — | — | — | |
| BiseNet* 18Speed (ms/f)=202020.04 | 73.8 | — | — | — | — | — | — | |
| GRFP(5)Speed (ms/f)=255, Max Latency (ms)=2552020.04 | 73.6 | — | — | — | — | — | — | |
| GRFP2021.02 | 73.6 | — | — | — | — | — | — | |
| LadderNetSpeed (ms/f)=332020.04 | 72.8 | — | — | — | — | — | — | |
| LadderNetSpeed (ms/f)=332020.04 | 72.8 | — | — | — | — | — | — | |
| MRCFABackbone=MiT-B0, Params (M)=4.2, FPS (f/s)=33.32022.07 | 72.8 | — | — | — | — | — | — | |
| MRCFABackbone=MiT-B02023.10 | 72.8 | — | — | 77.5 | 4.2 | 16.55 | — | |
| MRCFABackbone=MiT-B0, Input size=512 x 10242026.03 | 72.8 | — | — | 77.5 | 4.2 | 16.6 | — | |
| AccelSpeed (fps)=2.9, Avg RT (ms)=340, MaxLatency (ms)=5752020.07 | 72.1 | — | — | — | — | — | — | |
| AccelBackbone=ResNet-101, FPS (f/s)=3.62022.07 | 72.1 | — | — | — | — | — | — | |
| AccelBackbone=R1012023.10 | 72.1 | — | — | 824.4 | — | — | — | |
| SegFormerBackbone=MiT-B0, Params (M)=3.7, FPS (f/s)=58.52022.07 | 71.9 | — | — | — | — | — | — | |
| SegFormerBackbone=MiT-B0, Input size=512 x 10242026.03 | 71.9 | — | — | 54.6 | 3.7 | 58.5 | — | |
| ETCBackbone=ResNet-18, Params (M)=13.2, FPS (f/s)=9.52022.07 | 71.1 | — | — | — | — | — | — | |
| ETCBackbone=R182023.10 | 71.1 | — | — | 434.1 | — | — | — | |
| DVSNBackbone=ResNet-101, FPS (f/s)=19.82022.07 | 70.3 | — | — | — | — | — | — | |
| DVSNBackbone=R1012023.10 | 70.3 | — | — | 978.4 | — | — | — | |
| PSPNetBackbone=MobileNetV2, Params (M)=13.7, FPS (f/s)=11.22022.07 | 70.2 | — | — | — | — | — | — | |
| Deep Fea. Flow2018.04 | 70.1 | 273 | 654 | — | — | — | — | |
| GRFP(5)interval between keyframes=52018.04 | 69.4 | 470 | 470 | — | — | — | — | |
| GRFPBackbone=ResNet-101, FPS (f/s)=3.22022.07 | 69.4 | — | — | — | — | — | — | |
| DFFSpeed (ms/f)=156, Max Latency (ms)=5752020.04 | 69.2 | — | — | — | — | — | — | |
| DFFSpeed (fps)=5.7, Avg RT (ms)=175, MaxLatency (ms)=6442020.07 | 69.2 | — | — | — | — | — | — | |
| DFFGFLOPs=>9192021.02 | 69.2 | — | — | — | — | — | — | |
| DFFBackbone=ResNet-101, FPS (f/s)=9.72022.07 | 68.7 | — | — | — | — | — | — | |
| DFFBackbone=R1012023.10 | 68.7 | — | — | 100.8 | — | — | — | |
| Clockwork NetBackbone=ResNet-101, interval between keyframes=52018.04 | 67.7 | 141 | 360 | — | — | — | — | |
| ICNetSpeed (ms/f)=202020.04 | 67.7 | — | — | — | — | — | — | |
| ICNetSpeed (ms/f)=202020.04 | 67.7 | — | — | — | — | — | — | |
| CCBackbone=VGG-16, FPS (f/s)=16.52022.07 | 67.7 | — | — | — | — | — | — | |
| CLKSpeed (ms/f)=158, Max Latency (ms)=1982020.04 | 64.4 | — | — | — | — | — | — | |
| ClockworkSpeed (fps)=5.6, Avg RT (ms)=177, MaxLatency (ms)=2212020.07 | 64.4 | — | — | — | — | — | — | |
| DVSNetSpeed (ms/f)=332020.04 | 63.2 | — | — | — | — | — | — | |
| DVSNetSpeed (ms/f)=332020.04 | 63.2 | — | — | — | — | — | — | |
| DVSNet-fastSpeed (fps)=30.4, Avg RT (ms)=332020.07 | 63.2 | — | — | — | — | — | — | |
| FCNBackbone=MobileNetV2, Params (M)=9.8, FPS (f/s)=14.22022.07 | 61.5 | — | — | — | — | — | — | |
| DiTTAPre-trained on=VSPW2026.04 | 46.9 | — | — | — | — | — | 77.9 |