AI Paper Atlas

Research map / Video / Video-language understanding

Self-supervised Pre-training and Contrastive Representation Learning for Multiple-choice Video QA

Seonhoon Kim, Seohyeong Jeong, Eunbyul Kim, Inho Kang, Nojun Kwak

AAAI 2021 · Video › Video-language understanding · approach: Video question answering · 33 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 locally aligned attention instead of globally aligned attention (ablation)Video QA model · TVQA+ validationQA accuracy: 71.62 $\pm$ 0.45 -> 72.29 $\pm$ 0.31Improved
add temporal localization span loss (ablation)Video QA model with globally aligned attention · TVQA+ validationQA accuracy: 71.62 $\pm$ 0.45 -> 73.45 $\pm$ 0.31Improved
add temporal localization span loss (ablation)Video QA model with locally aligned attention · TVQA+ validationQA accuracy: 72.29 $\pm$ 0.31 -> 73.53 $\pm$ 0.29Improved
add multiple token type embeddings (ablation)Video QA model with global attention and temporal localization · TVQA+ validationQA accuracy: 73.45 $\pm$ 0.31 -> 73.98 $\pm$ 0.27Improved
add multiple token type embeddings (ablation)Video QA model with local attention and temporal localization · TVQA+ validationQA accuracy: 73.53 $\pm$ 0.29 -> 74.54 $\pm$ 0.21Improved
add supervised contrastive loss using a masked ground-truth answer as anchor (ablation)Video QA model with locally aligned attention, TL, and MT · TVQA+ validationQA accuracy: 74.54\% to 75.16\%Improved
add self-supervised question-prediction pre-training (ablation)Video QA model with TL, MT, and contrastive loss · TVQA+ validationQA accuracy: 75.16 $\pm$ 0.18 -> 75.83 $\pm$ 0.06Improved

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