AI Paper Atlas

Research map / Video / Video-language understanding

UATVR: Uncertainty-Adaptive Text-Video Retrieval

Bo Fang, Wenhao Wu, Chang Liu, Yu Zhou, Yuxin Song, Weiping Wang, Xiangbo Shu, Xiangyang Ji, Jingdong Wang

ICCV 2023 · Video › Video-language understanding · approach: Text-video retrieval · 95 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add dynamic semantic adaptation learnable tokens to token-wise baseline (ablation)token-wise interaction baseline · MSR-VTT (t2v, ViT-B/16)R@1: 48.4 -> 49.6Improved
add distribution-based uncertainty adaptation with KL constraint (ablation)TI + DSA · MSR-VTT (t2v, ViT-B/16)R@1: 49.6 -> 50.8Improved
use soft contrastive loss instead of multi-instance contrast (ablation)DUA · MSR-VTT (t2v, ViT-B/16)R@1: 50.8 -> 50.6Hurt
increase extra visual tokens from 3 to 4 or more (ablation)DSA · MSR-VTT (t2v, ViT-B/16)R@1: 49.6 -> 49.3 (Cv=4), 48.8 (Cv=8)Hurt
increase number of probabilistic embeddings K from 1 to 7 (ablation)DUA · MSR-VTT (t2v, ViT-B/16)R@1: 49.6 -> 50.8Improved
increase number of frames from 4 to 12 (ablation)UATVR · MSR-VTT (t2v, ViT-B/16)R@1: 44.9 -> 50.8Improved
increase frames from 12 to 16 or 20 (ablation)UATVR · MSR-VTT (t2v, ViT-B/16)R@1: 50.8 -> 51.0 (16), 50.8 (20)No clear effect
add inverted dual softmax post-processingUATVR · MSR-VTT (t2v, ViT-B/16)R@1: 50.8 -> 53.5Improved
UATVR (ViT-B/32) versus X-Pool on MSVDMSVD t2v · MSVD (t2v)R@1: 47.2 vs 46.0Hurt

Later papers that compared with this method

7 later accepted papers report UATVR as a baseline in their results tables; 7 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy