File size: 2,836 Bytes
d256d40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
71b4bed
d256d40
 
 
71b4bed
d256d40
 
 
71b4bed
d256d40
 
 
 
 
71b4bed
d256d40
 
 
 
 
 
 
 
 
 
71b4bed
d256d40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
71b4bed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
---
pretty_name: AV-QuantBench
license: cc-by-nc-4.0
task_categories:
- question-answering
- video-text-to-text
tags:
- multimodal
- audio
- video
- time-series
- synthetic
- benchmark
size_categories:
- 10K<n<100K
annotations_creators:
- machine-generated
language_creators:
- machine-generated
language:
- en
multilinguality:
- monolingual
source_datasets:
- original
---

# AV-QuantBench

AV-QuantBench is a procedural audio-visual benchmark for evaluating multimodal foundation models on abstract temporal reasoning, cross-modal conflict detection, and synchronized data interpretation across finance, medical, and industrial domains.

This Hugging Face dataset repository is structured as a benchmark-style release. It contains:

- split metadata in JSONL format,
- question-answer annotations,
- audio-visual sample assets,
- manifest files by domain,
- and documentation for schema and responsible use.

The repository is organized so that newly generated benchmark outputs can be added with minimal restructuring. In particular, the `samples/` directory follows the same domain-first layout as the AV-QuantBench generator outputs.

## Dataset Summary

AV-QuantBench converts time-series signals into synchronized visual topology and acoustic momentum. Each sample is paired with machine-generated QA derived from deterministic state-machine triggers. The benchmark is designed to evaluate whether a model can jointly reason over audio and video when the two modalities either align or intentionally diverge.

### Supported Tasks

- Cross-modal conflict detection
- Audio-visual temporal reasoning
- Multimodal question answering on synthetic data videos
- Robustness evaluation under counterfactual splicing

### Modalities

- Video (`.mp4`)
- Audio (`.wav`)
- Structured annotations (`.json`, `.jsonl`)

### Domains

- Finance
- Medical monitoring
- Industrial IoT

## Repository Structure

```text
AV-QuantBench-HF-Dataset/
├── README.md
├── LICENSE
├── .gitattributes
├── dataset_infos.json
├── metadata/
│   ├── train.jsonl
│   ├── val.jsonl
│   └── test.jsonl
├── qa/
│   ├── train.jsonl
│   ├── val.jsonl
│   └── test.jsonl
├── manifests/
│   ├── finance.jsonl
│   ├── medical.jsonl
│   └── iiot.jsonl
├── samples/
│   ├── finance/
│   │   ├── videos/
│   │   ├── audio/
│   │   ├── qa/
│   │   └── metadata/
│   ├── medical/
│   │   ├── videos/
│   │   ├── audio/
│   │   ├── qa/
│   │   └── metadata/
│   └── iiot/
│       ├── videos/
│       ├── audio/
│       ├── qa/
│       └── metadata/
└── docs/
    ├── schema.md
    └── responsible_use.md