Natural Language Queries on Ego4D NLQ (test)
26.67R@1 (IoU=0.3)GroundVQA
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GroundVQAtraining_mode=pre-trained solely on the VLG task and further fine-tuned on NLQv22023.12 | 26.67 | 39.94 | 17.63 | 27.7 | 22.15 | |
| GroundNLQ2023.12 | 24.5 | 40.46 | 17.31 | 29.17 | 20.91 | |
| GroundVQABtraining_mode=simultaneously trained on all three tasks with QAEGO4D and EGOTIMEQA data2023.12 | 23.65 | 36.19 | 14.96 | 24.58 | 19.31 | |
| NaQ++Architecture=ReLER, Features=InternVideo, Leaderboard Entry=CVPR '232023.01 | 21.7 | 25.12 | 13.64 | 16.33 | 17.67 | |
| NaQ++2023.12 | 21.7 | 25.12 | 13.64 | 16.33 | 17.67 | |
| NaQ2023.01 | 18.46 | 21.5 | 10.74 | 13.74 | 14.59 | |
| InternVideo2023.01 | 16.46 | 22.95 | 10.06 | 16.11 | 13.26 | |
| Badgers@UW-Madison2023.01 | 15.71 | 28.45 | 9.57 | 18.03 | 12.64 | |
| CONE2023.01 | 15.26 | 26.42 | 9.24 | 16.51 | 12.25 | |
| ReLER2023.01 | 12.89 | 15.41 | 8.14 | 9.94 | 10.51 | |
| ReLER2023.12 | 12.89 | 15.41 | 8.14 | 9.94 | 10.51 | |
| VSLNetVis-text Enc=Frozen+EgoNCE, Vis-text PT=EgoClip2022.06 | 10.46 | 16.76 | 6.24 | 11.29 | — | |
| EgoVLP2023.01 | 10.46 | 16.76 | 6.24 | 11.29 | 8.35 | |
| EgoVLP2023.12 | 10.46 | 16.76 | 6.24 | 11.29 | 8.35 | |
| VSLNetVis-text Enc=Frozen, Vis-text PT=EgoClip2022.06 | 10.34 | 15.81 | 6.24 | 10.39 | — | |
| VSLNetVis-text Enc=SlowFast+BERT2022.06 | 5.47 | 11.21 | 2.8 | 6.57 | — | |
| VSLNet2023.01 | 5.42 | 8.79 | 2.75 | 5.07 | 4.08 | |
| VSLNet2023.12 | 5.42 | 8.79 | 2.75 | 5.07 | 4.08 | |
| VSLNetVis-text Enc=Frozen, Vis-text PT=CC3M+WebVid-2M2022.06 | 4.87 | 8.67 | 2.5 | 4.97 | — | |
| VSLNetVis-text Enc=Frozen, Vis-text PT=HowTo100M2022.06 | 3.77 | 6.87 | 1.62 | 3.45 | — | |
| Ensembleconstituent_models=Ours-full-slowfast + Ours-full-omnivore, training_data=train set + val set, ensemble_strategy=top-5 prediction score sorting2022.07 | 0.1289 | 0.1541 | 0.0814 | 0.0994 | — |