1
stars
2
commits
2
linked in READMEs
Jun 1, 2026
updated
README
LVOmniBench
This repository provides the question-answering metadata of the LVOmniBench benchmark in a format compatible with lmms-eval.
The data is provided as a single parquet file containing only the QA annotations. Since raw videos are not included, please download them from the original release and match them with the QA annotations using video_id.
Task configurations and evaluation scripts are available in the SEATS repository: https://github.com/xxayt/SEATS.
Overview
LVOmniBench is a benchmark for long-form audio-visual understanding with Omnimodal LLMs. Videos are 10–90 minutes long (≈ 34 min on average), about 6× longer than typical audio-visual benchmarks, and every question requires joint reasoning across audio and visual modalities.
| Statistic | Value |
|---|---|
| QA pairs | 1,014 (multiple choice, 4 options) |
| Videos | 10–90 minutes, average ≈ 2,069 s (~34 min) |
| Audio types | Speech, Sound, Music, Mixed |
| Difficulty | Low / Medium / High |
| Question types | Human-Centric Understanding, Event Reasoning, Causal, ... |
| Language | English |
| License | Apache-2.0 |
License
Released under the Apache-2.0 license. The upstream KD-TAO/LVOmniBench release does not specify a license, so this repackaged QA split defaults to Apache-2.0.
Contributors
2 commits
Similar repos
1
stars
2
commits
2
linked in READMEs
Jun 1, 2026
updated
README
LVOmniBench
This repository provides the question-answering metadata of the LVOmniBench benchmark in a format compatible with lmms-eval.
The data is provided as a single parquet file containing only the QA annotations. Since raw videos are not included, please download them from the original release and match them with the QA annotations using video_id.
Task configurations and evaluation scripts are available in the SEATS repository: https://github.com/xxayt/SEATS.
Overview
LVOmniBench is a benchmark for long-form audio-visual understanding with Omnimodal LLMs. Videos are 10–90 minutes long (≈ 34 min on average), about 6× longer than typical audio-visual benchmarks, and every question requires joint reasoning across audio and visual modalities.
| Statistic | Value |
|---|---|
| QA pairs | 1,014 (multiple choice, 4 options) |
| Videos | 10–90 minutes, average ≈ 2,069 s (~34 min) |
| Audio types | Speech, Sound, Music, Mixed |
| Difficulty | Low / Medium / High |
| Question types | Human-Centric Understanding, Event Reasoning, Causal, ... |
| Language | English |
| License | Apache-2.0 |
License
Released under the Apache-2.0 license. The upstream KD-TAO/LVOmniBench release does not specify a license, so this repackaged QA split defaults to Apache-2.0.
Similar repos
Linked in READMEs
Contributors
2 commits