Papers › Video-adverb retrieval with compositional adverb-action embeddings

Video-adverb retrieval with compositional adverb-action embeddings

26 Sep 2023arXiv:2309.15086archive 2025-07-28

Thomas Hummel, Otniel-Bogdan Mercea, A. Sophia Koepke, Zeynep Akata

Retrieving adverbs that describe an action in a video poses a crucial step towards fine-grained video understanding. We propose a framework for video-to-adverb retrieval (and vice versa) that aligns video embeddings with their matching compositional adverb-action text embedding in a joint embedding space. The compositional adverb-action text embedding is learned using a residual gating mechanism, along with a novel training objective consisting of triplet losses and a regression target. Our method achieves state-of-the-art performance on five recent benchmarks for video-adverb retrieval. Furthermore, we introduce dataset splits to benchmark video-adverb retrieval for unseen adverb-action compositions on subsets of the MSR-VTT Adverbs and ActivityNet Adverbs datasets. Our proposed framework outperforms all prior works for the generalisation task of retrieving adverbs from videos for unseen adverb-action compositions. Code and dataset splits are available at https://hummelth.github.io/ReGaDa/.

PaperPDFCode

Code

ExplainableML/ReGaDa officialmentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Video-Adverb RetrievalVideo-Adverb Retrieval (Unseen Compositions)

1 archive task tag without a task page not shown.

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Video-Adverb Retrieval AIR ReGaDa Acc-A 0.874 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval AIR ReGaDa mAP M 0.418 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval AIR ReGaDa mAP W 0.704 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval ActivityNet Adverbs ReGaDa Acc-A 0.771 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval ActivityNet Adverbs ReGaDa mAP M 0.175 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval ActivityNet Adverbs ReGaDa mAP W 0.239 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval HowTo100M Adverbs ReGaDa Acc-A 0.817 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval HowTo100M Adverbs ReGaDa mAP M 0.528 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval HowTo100M Adverbs ReGaDa mAP W 0.567 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval MSR-VTT Adverbs ReGaDa Acc-A 0.786 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval MSR-VTT Adverbs ReGaDa mAP M 0.228 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval MSR-VTT Adverbs ReGaDa mAP W 0.378 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval VATEX Adverbs ReGaDa Acc-A 0.817 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval VATEX Adverbs ReGaDa mAP M 0.113 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval VATEX Adverbs ReGaDa mAP W 0.29 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) ActivityNet Adverbs ReGaDa Acc-A 58.4 #1 of 4 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) ActivityNet Adverbs Pseudo Adverbs Acc-A 57 #2 of 4 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) ActivityNet Adverbs CLIP Acc-A 55.1 #3 of 4 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) ActivityNet Adverbs Action Changes (reg) Acc-A 53.9 #4 of 4 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) MSR-VTT Adverbs ReGaDa Acc-A 61 #1 of 5 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) MSR-VTT Adverbs Action Changes (reg) Acc-A 59 #2 of 5 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) MSR-VTT Adverbs CLIP Acc-A 57 #3 of 5 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) MSR-VTT Adverbs Pseudo Adverbs Acc-A 56 #4 of 5 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) MSR-VTT Adverbs Action Changes (cls) Acc-A 53.7 #5 of 5 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) VATEX Adverbs ReGaDa Acc-A 61.7 #1 of 5 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) VATEX Adverbs Action Changes (reg) Acc-A 54.9 #2 of 5 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) VATEX Adverbs CLIP Acc-A 54.5 #3 of 5 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) VATEX Adverbs Action Changes (cls) Acc-A 54.3 #4 of 5 Archive leaderboard report
Video-Adverb Retrieval (Unseen Compositions) VATEX Adverbs Pseudo Adverbs Acc-A 53.8 #5 of 5 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Introduced by this paper: ReGaDa

ReGaDa

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections