AI Paper Atlas

Research map / Video / Video-language understanding

COOT: Cooperative Hierarchical Transformer for Video-Text Representation Learning

Simon Ging, Mohammadreza Zolfaghari, Hamed Pirsiavash, Thomas Brox

NeurIPS 2020 · Video › Video-language understanding · approach: Video-language pretraining · 74 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
Use COOT instead of HSEActivityNet-captions video-paragraph retrieval · ActivityNet-captions val1, paragraph→videoR@1: 44.4±0.5 -> 60.8±0.6Improved
Use COOT instead of HSEActivityNet-captions video-paragraph retrieval · ActivityNet-captions val2, paragraph→videoR@1: 32.9 -> 48.5Improved
Use COOT with HowTo100M-pretrained video-text features instead of MIL-NCEYouCook2 paragraph-to-video retrieval · YouCook2, paragraph→videoR@1: 61.9* -> 77.2±1.0Improved
Add attention-aware feature aggregation rather than max pooling (ablation)COOT · ActivityNet-captions val1, paragraph→videoR@1: 58.2±0.5 -> 59.0±0.5Improved
Add cross-modal cycle-consistency loss (ablation)COOT with AFA and CoT · ActivityNet-captions val1, paragraph→videoR@1: 59.8±1.1 -> 60.8±0.6Improved
Add the contextual transformer (ablation)COOT with AFA and CMC · ActivityNet-captions val1, paragraph→videoR@1: 59.5±0.5 -> 60.8±0.6Improved
Replace BERT with GloVe text features (ablation)COOT · ActivityNet-captions val1, paragraph→videoR@1: 60.8±0.6 -> 56.5±1.1Hurt
Remove the low-level alignment loss (ablation)COOT full objective · ActivityNet-captions val1, paragraph→videoR@1: 60.8±0.6 -> 58.5±1.1Hurt
Add global-context clustering loss (ablation)COOT · ActivityNet-captions val1, paragraph→videoR@1: 60.8±0.6 -> 60.8±0.6No clear effect
Use attention-aware feature aggregation instead of average pooling at the high level (ablation)COOT · ActivityNet-captions val1, paragraph→videoR@1: 60.8±0.6 -> 42.6±0.4Hurt
Use two high-level attention-aware feature aggregation modules instead of one (ablation)COOT · ActivityNet-captions val1, paragraph→videoR@1: 42.6±0.4 -> 42.8±0.1No clear effect
Use HowTo100M-pretrained features instead of classification-pretrained features (ablation)COOT on YouCook2 · YouCook2, paragraph→videoR@1: 50.4±2.6 -> 77.2±1.0Improved

Later papers that compared with this method

1 later accepted paper reports COOT as a baseline in their results tables; 1 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 0 of 1 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