Video-to-Text Retrieval on VATEX (test)
82.3Recall@1Side4Video (L/14)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Side4Video (L/14)Protocol=Frozen backbone, Pretrain=CLIP-400M, Backbone=ViT-L/142023.11 | 82.3 | 99.4 | 99.9 | 1 | 1.3 | |
| Side4Video (E/14)Protocol=Frozen backbone, Pretrain=LAION-2B, Backbone=ViT-E/142023.11 | 79.7 | 98.9 | 99.8 | 1 | 1.4 | |
| Side4Video (B/16)Protocol=Frozen backbone, Pretrain=CLIP-400M, Backbone=ViT-B/162023.11 | 79 | 98 | 99.5 | 1 | 1.5 | |
| Side4Video (B/32)Protocol=Frozen backbone, Pretrain=CLIP-400M, Backbone=ViT-B/322023.11 | 77.2 | 97.6 | 99.1 | 1 | 1.6 | |
| CLIP4Clip (ViT)Protocol=Frozen backbone, Pretrain=CLIP-400M2023.11 | 76.3 | 98 | 99.2 | 1 | 1.6 | |
| CLIP2Video2021.06 | 76 | 97.7 | 99.9 | 1 | 1.5 | |
| CM AdapterProtocol=Frozen backbone, Pretrain=CLIP-400M2023.11 | 74.7 | 97.2 | 99.1 | — | 1.6 | |
| VideoCoCazero-shot=true2022.12 | 73.6 | 93.2 | 97.2 | — | — | |
| CLIP4Clip-seqTransf2021.06 | 73.2 | 97.1 | 99.1 | 1 | 1.7 | |
| InternVideozero-shot=true2022.12 | 69.5 | — | — | — | — | |
| VideoCoCa (w.o. videos)zero-shot=true, initialization=pretrained CoCa model without continued pretraining2022.12 | 64.1 | 90.2 | 95.6 | — | — | |
| CoCazero-shot=true2022.12 | 60.9 | 88.1 | 92.4 | — | — | |
| SUPPORT-SET2021.06 | 58.4 | 84.4 | 91 | 1 | — | |
| CLIP2021.06 | 52.7 | 88.8 | 94.9 | 1 | 3.8 | |
| Random Baseline2021.06 | 0.02 | 0.1 | 1.02 | 2,100.5 | — |