AVSD
Harbor: pendingThe Audio Visual Scene-Aware Dialog (AVSD) dataset, or DSTC7 Track 3, is a audio-visual dataset for dialogue understanding.
- Samples
- 14,188
- Released
- 2018
- Scores
- 0
AVSD-Scenes
AVSD-Scenes: A Dataset for Audio-Visual Description of Urban Scenes
A dataset containing 12,291 paired audio-visual scene descriptions generated from the TAU Urban Audio-Visual Scenes dataset using large language models.
How to run
ltb · avsd-scenes
$ npm i -g lettokenburn-cli$ ltb bench info avsd-scenes
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
3 sourced entriesEvery number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
The AVSD-Scenes dataset consists of natural-language descriptions for urban scenes derived from synchronized audio and visual recordings. It is constructed in two stages: first generating modality-specific captions using Qwen2-Audio-7B-Instruct and Qwen2.5-VL-7B-Instruct, then fusing them with LLMs like Qwen3 or Mistral to create unified descriptions.
The dataset supports benchmarks for cross-modal retrieval, semantic alignment, scene classification, and description quality evaluation using both automated metrics (accuracy) and human subjective assessment.