Action Step Localization on CrossTask
47.3Average RecallVideoCLIP
Evaluation Results
| Method | Links | |
|---|---|---|
| VideoCLIPEvaluation Protocol=Fine-tuned2021.09 | 47.3 | |
| TACo2021.08 | 42.5 | |
| UniVLEvaluation Protocol=Supervised2021.09 | 42 | |
| UniVL(v3)2021.08 | 42 | |
| UniVLBackbone=S3D-G, Zero-Shot=true2022.12 | 42 | |
| ActBERTEvaluation Protocol=Supervised2021.09 | 41.4 | |
| ActBert2021.08 | 41.4 | |
| ActBERTBackbone=+ Faster-RCNN, Zero-Shot=true2022.12 | 41.4 | |
| MMFTBackbone=S3D-G, Zero-Shot=true2022.12 | 41.1 | |
| MIL-NCEEvaluation Protocol=Zero-shot2021.09 | 40.5 | |
| MIL-NCE2021.08 | 40.5 | |
| MIL-NCEBackbone=S3D-G, Zero-Shot=true2022.12 | 40.5 | |
| MMFTBackbone=R152 + RX101, Zero-Shot=true2022.12 | 39.3 | |
| ActBERTBackbone=R101 + Res3D, Zero-Shot=true2022.12 | 37.1 | |
| TempCLRBackbone=S3D-G, Zero-Shot=true2022.12 | 36.9 | |
| MIL-NCEBackbone=I3D-G, Zero-Shot=true2022.12 | 36.4 | |
| MCNBackbone=R152 + RX101, Zero-Shot=true2022.12 | 35.1 | |
| VideoCLIPEvaluation Protocol=Zero-shot2021.09 | 33.9 | |
| VideoCLIPBackbone=S3D-G, Zero-Shot=true2022.12 | 33.9 | |
| HowTo100MEvaluation Protocol=Zero-shot2021.09 | 33.6 | |
| TVJE2021.08 | 33.6 | |
| HT100MBackbone=R152 + RX101, Zero-Shot=true2022.12 | 33.6 | |
| SupervisedEvaluation Protocol=Supervised2021.09 | 31.6 | |
| Supervised2021.08 | 31.6 | |
| ZhukovEvaluation Protocol=Supervised2021.09 | 22.4 | |
| Zhukov et al.2021.08 | 22.4 | |
| AlayracEvaluation Protocol=Supervised2021.09 | 13.3 | |
| Alayrac et al.2021.08 | 13.3 |