Continuous Sign Language Recognition on PHOENIX14-T (dev)
17WERDCA
Evaluation Results
| Method | Links | |
|---|---|---|
| DCAextra cues (keypoints)=false2023.05 | 17 | |
| CorrNet+Venue=arXiv 20242025.12 | 17.2 | |
| C2STextra cues (keypoints)=false2023.05 | 17.3 | |
| USTMbaseModality=RGB2025.12 | 17.4 | |
| USTMsmallModality=RGB2025.12 | 17.4 | |
| USTMsmallBackbone=Swin-small2025.12 | 17.4 | |
| USTMsmallVariant=small2025.12 | 17.4 | |
| USTMbaseModality=RGB2025.12 | 17.6 | |
| ACMCVenue=IMAVIS 20252025.12 | 17.6 | |
| USTMbaseBackbone=Swin-base2025.12 | 17.6 | |
| USTMbaseVariant=base2025.12 | 17.6 | |
| TwoStream-SLRextra cues (keypoints)=true2023.05 | 17.7 | |
| TwoStream-SLRModality=RGB+Pose2025.12 | 17.7 | |
| SlowFastSignVenue=ICASSP 20242025.12 | 17.7 | |
| USTMtinyModality=RGB2025.12 | 17.8 | |
| USTMtinyBackbone=Swin-tiny2025.12 | 17.8 | |
| USTMtinyVariant=tiny2025.12 | 17.8 | |
| USTMsmallModality=RGB2025.12 | 17.9 | |
| SlowFastSign (Reproduced)Venue=ICASSP 2024, reproducibility=Reproduced2025.12 | 17.9 | |
| USTMtinyModality=RGB2025.12 | 18.1 | |
| TwoStream-SLRModality=RGB+Pose2025.12 | 18.4 | |
| Swin-MSTPVenue=Neurocomp. 2025, Backbone=Swin-small2025.12 | 18.7 | |
| SignBERT+extra cues (keypoints)=false2023.05 | 18.8 | |
| CorrNetextra cues (keypoints)=false2023.05 | 18.9 | |
| CorrNetVenue=CVPR 20232025.12 | 18.9 | |
| STNetVenue=IEEE Acc. 20252025.12 | 19.2 | |
| SENBackbone=ResNet18, GFLOPS per video=3672023.08 | 19.3 | |
| AdaBrowsealpha=0.04, GFLOPS=287, Throughput (videos/s)=12.312023.08 | 19.3 | |
| SENextra cues (keypoints)=false2023.05 | 19.3 | |
| CTCAextra cues (keypoints)=false2023.05 | 19.3 | |
| SENVenue=AAAI 20232025.12 | 19.3 | |
| CTCAVenue=CVPR 20232025.12 | 19.3 | |
| TLPBackbone=ResNet18, GFLOPS per video=3672023.08 | 19.4 | |
| AdaBrowseBackbone=ResNet18, GFLOPS per video=2542023.08 | 19.4 | |
| AdaBrowsealpha=0.15, GFLOPS=254, Throughput (videos/s)=14.262023.08 | 19.4 | |
| AdaBrowse+alpha=0.05, GFLOPS=241, Throughput (videos/s)=15.122023.08 | 19.4 | |
| TLPextra cues (keypoints)=false2023.05 | 19.4 | |
| CVT-SLRextra cues (keypoints)=false2023.05 | 19.4 | |
| CVT-SLRModality=RGB+Pose2025.12 | 19.4 | |
| TLPVenue=ECCV 20222025.12 | 19.4 | |
| AdaBrowse+Backbone=ResNet18, GFLOPS per video=1712023.08 | 19.5 | |
| BaselineGFLOPS=364, Throughput (videos/s)=12.222023.08 | 19.5 | |
| AdaBrowse+alpha=0.1, GFLOPS=171, Throughput (videos/s)=17.582023.08 | 19.5 | |
| CoSignextra cues (keypoints)=false2023.05 | 19.5 | |
| AdaBrowseVenue=ACM MM 20232025.12 | 19.5 | |
| CoSignPre-Training=false, Encoder Params=≈ 10M, Decoder Params=≈ 18M2026.03 | 19.5 | |
| STMCBackbone=VGG11, GFLOPS per video=1554, Extra Modalities=Face and Hands2023.08 | 19.6 | |
| STMCEnd-to-end=false, Modalities (Training)=R+P, Modalities (Inference)=R2022.12 | 19.6 | |
| STMCModality=RGB+Pose2025.12 | 19.6 | |
| CVT-SLRModality=RGB+Pose2025.12 | 19.8 | |
| CLIP-SLAVenue=CVPR 20252025.12 | 19.8 | |
| SignBERTModality=RGB+Pose2025.12 | 20.1 | |
| MP-SLRVenue=TPAMI 20252025.12 | 20.1 | |
| MSKAVenue=PR 20252025.12 | 20.1 | |
| MSKAPre-Training=false, Encoder Params=≈ 15M, Decoder Params=≈ 28M2026.03 | 20.1 | |
| C2SLRBackbone=ResNet18, GFLOPS per video=367, Extra Modalities=Face and Hands2023.08 | 20.2 | |
| C2SLREnd-to-end=true, Modalities (Training)=R+P, Modalities (Inference)=R2022.12 | 20.2 | |
| C2SLRextra cues (keypoints)=true2023.05 | 20.2 | |
| C2SLRModality=RGB+Pose2025.12 | 20.2 | |
| C2SLRModality=RGB+Pose2025.12 | 20.5 | |
| SSSLRVenue=ICASSP 20232025.12 | 20.5 | |
| STMCModality=RGB+Pose2025.12 | 20.6 | |
| SMKDBackbone=ResNet18, GFLOPS per video=3672023.08 | 20.8 | |
| SMKDEnd-to-end=false, Modalities (Training)=R, Modalities (Inference)=R2022.12 | 20.8 | |
| SMKDVenue=ICCV 20212025.12 | 20.8 | |
| STARKPre-Training=false, Encoder Params=≈ 3M, Decoder Params=≈ 28M2026.03 | 21 | |
| MMTLBEnd-to-end=true, Modalities (Training)=R, Modalities (Inference)=R2022.12 | 21.9 | |
| V-L Mapperextra cues (keypoints)=false2023.05 | 21.9 | |
| CNN+HMM+LSTMBackbone=GoogLeNet, GFLOPS per video=316, Extra Modalities=Face and Hands2023.08 | 22.1 | |
| CNN-LSTM-HMMsEnd-to-end=false, Modalities (Training)=R, Modalities (Inference)=R2022.12 | 22.1 | |
| DNFModality=RGB+Op.Flow2025.12 | 23.1 | |
| FCNBackbone=Custom, GFLOPS per video=2862023.08 | 23.3 | |
| FCNEnd-to-end=true, Modalities (Training)=R, Modalities (Inference)=R2022.12 | 23.3 | |
| FCNVenue=ECCV 20202025.12 | 23.3 | |
| CMAVenue=ACM MM 20202025.12 | 24.1 | |
| SLTBackbone=GoogLeNet, GFLOPS per video=316, Extra Modalities=Face and Hands2023.08 | 24.5 | |
| SLTEnd-to-end=true, Modalities (Training)=R, Modalities (Inference)=R2022.12 | 24.6 | |
| SFLEnd-to-end=true, Modalities (Training)=R, Modalities (Inference)=R2022.12 | 25.1 | |
| TwoStream-SLRPre-Training=true2026.03 | 27.1 | |
| SignBERT+Pre-Training=true2026.03 | 32.9 |