AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

OmniVL: One Foundation Model for Image-Language and Video-Language Tasks

Junke Wang, Dongdong Chen, Zuxuan Wu, Chong Luo, Luowei Zhou, Yucheng Zhao, Yujia Xie, Ce Liu, Yu-Gang Jiang, Lu Yuan

NeurIPS 2022 · Vision-language and multimodal › Human and embodied modeling · approach: Vision-language pretraining · 6 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
Decoupled joint pretraining instead of joint pretraining from scratch (ablation)OmniVL · MSRVTT text-to-video retrievalIR@1: 23.6 -> 47.8Improved
Decoupled joint pretraining instead of joint pretraining from scratch (ablation)OmniVL · COCO retrievalTR@1: 50.2 -> 82.1Improved
Decoupled joint pretraining instead of image-only pretraining (ablation)OmniVL · MSRVTT text-to-video retrievalIR@1: 38.2 -> 47.8Improved
Decoupled joint pretraining instead of image-only pretraining (ablation)OmniVL · COCO retrievalTR@1: 80.9 -> 82.1Improved
Decoupled joint pretraining instead of Img2Vid pretraining (ablation)OmniVL · MSRVTT text-to-video retrievalIR@1: 42.5 -> 47.8Improved
Img2Vid pretraining instead of image-only pretraining (ablation)OmniVL · COCO retrievalTR@1: 80.9 -> 79.7Hurt
Joint pretraining from scratch instead of image-only pretraining (ablation)OmniVL · COCO captionCIDEr: 131.6 -> 94.6Hurt
Video-only pretraining instead of no pretraining (ablation)OmniVL · MSRVTT-QAaccuracy: 36.6 -> 15.8Hurt
Joint pretraining from scratch instead of video-only pretraining (ablation)OmniVL · MSRVTT text-to-video retrievalIR@1: 13.7 -> 23.6Improved
Image-only pretraining versus no pretraining (ablation)OmniVL · MSRVTT text-to-video retrievalIR@1: 9.6 -> 38.2Improved
Replace UniVLC loss with vanilla contrastive loss (w/o visual-label data) on vision-language tasks (ablation)OmniVL · retrieval, captioning, VQAvariousNo clear effect
Replace UniVLC loss with vanilla contrastive loss on visual-only tasks (ablation)OmniVL · linear probing and action recognitionaccuracyImproved

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