Video-Text Retrieval on MSVD
78.1R@1MAVIS
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| MAVISMethod Category=Agentic Methods2026.06 | 78.1 | — | — | — | — | 94.25 | 97.33 | — | |
| MERLIN (Round 5)Method Category=Agentic Methods2026.06 | 77.61 | — | — | — | — | 94.48 | 97.31 | — | |
| MERLIN (Round 3)Method Category=Agentic Methods2026.06 | 71.79 | — | — | — | — | 91.79 | 96.87 | — | |
| UMT (Base)backbone=UMT-L, pretraining_pairs=17M, eval_protocol=zero-shot2024.03 | 62.7 | — | — | — | 267.8 | 83.7 | 89.7 | — | |
| vid-TLDRbackbone=UMT-L, pretraining_pairs=17M, eval_protocol=zero-shot2024.03 | 62.7 | — | — | — | 165.3 | 83.8 | 90.3 | — | |
| vid-TLDRbackbone=UMT-L, pretraining_pairs=25M, eval_protocol=zero-shot2024.03 | 62.5 | — | — | — | 153.6 | 83.2 | 88.8 | — | |
| UMT (Base)backbone=UMT-L, pretraining_pairs=25M, eval_protocol=zero-shot2024.03 | 61.8 | — | — | — | 267.8 | 83.3 | 88.8 | — | |
| UMT + ToMebackbone=UMT-L, pretraining_pairs=17M, eval_protocol=zero-shot2024.03 | 61 | — | — | — | 182.1 | 83.2 | 88.8 | — | |
| Marengo-2.6Method Category=Large-scale Video Foundation Models2026.06 | 61 | — | — | — | — | 85.3 | 91.5 | — | |
| UMT + ToMebackbone=UMT-L, pretraining_pairs=25M, eval_protocol=zero-shot2024.03 | 60.3 | — | — | — | 182.1 | 82.5 | 88.3 | — | |
| InternVideo2-6BMethod Category=Large-scale Video Foundation Models2026.06 | 59.3 | — | — | — | — | 84.4 | 89.6 | — | |
| VideoPrism-gMethod Category=Large-scale Video Foundation Models2026.06 | 58.5 | — | — | — | — | 83.4 | 89.2 | — | |
| vid-TLDRbackbone=UMT-L, pretraining_pairs=5M, eval_protocol=zero-shot2024.03 | 56.3 | — | — | — | 165.3 | 79.6 | 86.1 | — | |
| VASTMethod Category=Large-scale Video Foundation Models2026.06 | 55.4 | — | — | — | — | 82 | 88.5 | — | |
| UMT (Base)backbone=UMT-L, pretraining_pairs=5M, eval_protocol=zero-shot2024.03 | 55.3 | — | — | — | 267.8 | 79.4 | 85.9 | — | |
| BridgeFormerReranking (Matching Loss)=false2023.10 | 54.4 | — | — | — | — | 82.8 | 89.4 | 1 | |
| LanguageBind-HMethod Category=Large-scale Video Foundation Models2026.06 | 53.9 | — | — | — | — | 80.4 | 87.8 | — | |
| UMT + ToMebackbone=UMT-L, pretraining_pairs=5M, eval_protocol=zero-shot2024.03 | 53.6 | — | — | — | 182.1 | 78.9 | 85.1 | — | |
| vid-TLDRbackbone=UMT-B, pretraining_pairs=25M, eval_protocol=zero-shot2024.03 | 53.6 | — | — | — | 50.6 | 78.1 | 85.4 | — | |
| vid-TLDRbackbone=UMT-B, pretraining_pairs=17M, eval_protocol=zero-shot2024.03 | 53.3 | — | — | — | 50.2 | 77 | 84 | — | |
| T-MASSMethod Category=Supervised / Fine-tuned2026.06 | 53 | — | — | — | — | 81 | 88.5 | — | |
| MERLIN (Round 0)Method Category=Agentic Methods2026.06 | 52.39 | — | — | — | — | 77.16 | 84.78 | — | |
| UMT (Base)backbone=UMT-B, pretraining_pairs=25M, eval_protocol=zero-shot2024.03 | 52.1 | — | — | — | 78.5 | 77.1 | 84.7 | — | |
| UMT (Base)backbone=UMT-B, pretraining_pairs=17M, eval_protocol=zero-shot2024.03 | 52 | — | — | — | 78.5 | 75.7 | 83.6 | — | |
| Cap4VideoMethod Category=Supervised / Fine-tuned2026.06 | 51.8 | — | — | — | — | 80.8 | 88.3 | — | |
| InternVideoBackbone=UniformerV2, Video patch dropping=true2023.10 | 51 | — | — | — | — | 80.5 | 88.4 | 1 | |
| UMT + ToMebackbone=UMT-B, pretraining_pairs=17M, eval_protocol=zero-shot2024.03 | 50.7 | — | — | — | 62.8 | 74.8 | 83 | — | |
| InternVideoBackbone=ViT, Video patch dropping=true2023.10 | 50.4 | — | — | — | — | 79.4 | 88 | 1 | |
| X-CLIPMethod Category=Supervised / Fine-tuned2026.06 | 50.4 | — | — | — | — | 80.6 | 87.1 | — | |
| UMT + ToMebackbone=UMT-B, pretraining_pairs=25M, eval_protocol=zero-shot2024.03 | 50.1 | — | — | — | 66.2 | 76.1 | 84.1 | — | |
| InternVideoBackbone=UniformerV2, Video patch dropping=false2023.10 | 49.9 | — | — | — | — | 79.7 | 87.8 | 2 | |
| BridgeFormerZero-shot=true2023.10 | 48.4 | — | — | — | — | 76.4 | 85.8 | 2 | |
| vid-TLDRbackbone=UMT-B, pretraining_pairs=5M, eval_protocol=zero-shot2024.03 | 48.1 | — | — | — | 50.6 | 72.5 | 80.4 | — | |
| UMT (Base)backbone=UMT-B, pretraining_pairs=5M, eval_protocol=zero-shot2024.03 | 47.4 | — | — | — | 78.5 | 72.2 | 80.2 | — | |
| TeachCLIPMethod Category=Supervised / Fine-tuned2026.06 | 47.4 | — | — | — | — | 77.3 | 84.5 | — | |
| CLIP4ClipReranking (Matching Loss)=false2023.10 | 46.2 | — | — | — | — | 76.1 | 84.6 | 2 | |
| CLIP4ClipMethod Category=Supervised / Fine-tuned2026.06 | 46.2 | — | — | — | — | 76.1 | 84.6 | — | |
| UMT + ToMebackbone=UMT-B, pretraining_pairs=5M, eval_protocol=zero-shot2024.03 | 45.6 | — | — | — | 66.2 | 70.8 | 79.8 | — | |
| InternVideoZero-shot=true, WiSE-FT=true2023.10 | 44 | — | — | — | — | 72.7 | 82.5 | 2 | |
| InternVideo (ViT Backbone)Zero-shot=true, Backbone=ViT, WiSE-FT=true2023.10 | 39.2 | — | — | — | — | 66.9 | 76.2 | 2 | |
| FrozenZero-shot=true2023.10 | 33.7 | — | — | — | — | 64.7 | 76.3 | 3 | |
| FROZENReranking (Matching Loss)=false2023.10 | 33.7 | — | — | — | — | 64.7 | 76.3 | 3 | |
| NoiseEstZero-shot=true2023.10 | 13.7 | — | — | — | — | 35.7 | 47.7 | 12 | |
| Fully-FTArchitecture=CLIP4Clip2023.08 | — | 45.2 | 57.1 | 425.5 | — | — | — | — | |
| Fully-FTParams (M)=151.3, Mem (G)=48.82026.04 | — | 45.2 | 57.1 | 425.5 | — | — | — | — | |
| Fully-FTParams. (M)=151.3, Mem. (G)=48.82026.04 | — | 45.2 | 57.1 | 425.5 | — | — | — | — | |
| LSTArchitecture=CLIP4Clip2023.08 | — | 35.5 | 55.4 | 407.2 | — | — | — | — | |
| LSTParams (M)=11.2, Mem (G)=32.02026.04 | — | 35.5 | 55.4 | 407.2 | — | — | — | — | |
| LSTParams. (M)=11.2, Mem. (G)=32.02026.04 | — | 35.5 | 55.4 | 407.2 | — | — | — | — | |
| Ours†Params. (M)=10.1, Mem. (G)=14.52026.04 | — | 42 | 60.4 | 435.6 | — | — | — | — | |
| Ours‡Params. (M)=10.1, Mem. (G)=14.62026.04 | — | 42.1 | 60.7 | 435.8 | — | — | — | — | |
| Ours♠Params (M)=9.7, Mem (G)=14.2, QPS=43.522026.04 | — | 42.5 | 61.3 | 435.8 | — | — | — | — | |
| Ours♡Params (M)=9.7, Mem (G)=14.0, QPS=43.652026.04 | — | 42.3 | 60 | 435.7 | — | — | — | — | |
| Partially⇓Architecture=CLIP4Clip2023.08 | — | 37.4 | 52.4 | 406.4 | — | — | — | — | |
| SHERLParams (M)=9.6, Mem (G)=13.6, QPS=32.532026.04 | — | 40.9 | 60.2 | 429.7 | — | — | — | — | |
| SHERLParams. (M)=9.6, Mem. (G)=13.62026.04 | — | 40.9 | 60.2 | 429.7 | — | — | — | — | |
| UniPTArchitecture=CLIP4Clip2023.08 | — | 40.9 | 59.7 | 432.1 | — | — | — | — | |
| UniPTParams (M)=9.6, Mem (G)=13.6, QPS=27.032026.04 | — | 40.9 | 59.7 | 432.1 | — | — | — | — | |
| UniPTParams. (M)=9.6, Mem. (G)=13.62026.04 | — | 40.9 | 59.7 | 432.1 | — | — | — | — |