AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

TVLT: Textless Vision-Language Transformer

Zineng Tang, Jaemin Cho, Yixin Nie, Mohit Bansal

NeurIPS 2022 · Vision-language and multimodal › Cross-modal learning · approach: Audio-visual speech separation · 16 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 raw audio rather than ASR text with YTT-S pretrainingTVLT · MSRVTT testR@1: 19.3 -> 23.8Improved
Use raw audio rather than ASR text with YTT-S pretrainingTVLT · YouCook validationR@1: 26.3 -> 32.8Improved
Use raw audio rather than ASR text with YTT-S pretrainingTVLT · CrossTask validationR@1: 12.2 -> 15.3Improved
Use raw audio rather than ASR text with YTT-S pretrainingTVLT · PlacesAudioR@1: 54.3 -> 49.0Hurt
Use raw audio rather than ASR text with YTT-S pretrainingTVLT · VQAv2 test-devaccuracy: 63.2 -> 61.0Hurt
Use raw audio rather than ASR text with YTT-S pretrainingTVLT · CMU-MOSEI testA2: 76.6 -> 76.8Improved
Use raw audio queries rather than caption textTVLT · MSRVTTR@1: 22.0 -> 20.1Hurt
Remove ASR from the inference pipelineTVLT · 20s video / 8 frameslatency: 2916 -> 103Improved
Add speech-span-focused audio masking (ablation)TVLT with 16×16 audio patches · MSRVTTR@1: 21.7 -> 22.3Improved
Add speech-span-focused audio masking (ablation)TVLT with 16×16 audio patches · VQAv2accuracy: 57.8 -> 58.6Improved
Use 2×128 rather than 16×16 audio patches (ablation)TVLT with speech-focused masking · MSRVTTR@1: 22.3 -> 21.2Hurt
Use 2×128 rather than 16×16 audio patches (ablation)TVLT with speech-focused masking · VQAv2accuracy: 58.6 -> 59.2Improved

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