File size: 6,310 Bytes
49bfa75 baa5123 49bfa75 6080413 6c7d33a 6080413 49bfa75 f62f0dc 49bfa75 2da3d9e 49bfa75 f62f0dc 4abc32b | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 | ---
pretty_name: OpenVisTool-Bench
license: other
language:
- en
task_categories:
- visual-question-answering
- image-to-text
tags:
- multimodal
- tool-use
- agentic-vision
- visual-reasoning
- gui-grounding
- web-to-html
size_categories:
- n<1K
configs:
- config_name: chart
data_files:
- split: test
path: data/chart.jsonl
- config_name: table
data_files:
- split: test
path: data/table.jsonl
- config_name: visual_search
data_files:
- split: test
path: data/visual_search.jsonl
- config_name: gui_grounding
data_files:
- split: test
path: data/gui_grounding.jsonl
- config_name: web_to_html
data_files:
- split: test
path: data/web_to_html.jsonl
---
# OpenVisTool-Bench
OpenVisTool-Bench is a 559-example, five-domain benchmark for visual agents
that can call inspection and editing tools. All five configurations use the
`test` split. This benchmark accompanies [OpenVisTool: An Open Recipe for Synthesizing Instructive Visual Tool-Use Trajectories](https://arxiv.org/abs/2608.08557).
## Configurations
| Configuration | Domain | Items | Metadata |
|---|---|---:|---|
| `chart` | Chart reasoning | 110 | `data/chart.jsonl` |
| `table` | Table reasoning | 126 | `data/table.jsonl` |
| `visual_search` | Fine-grained visual search | 106 | `data/visual_search.jsonl` |
| `gui_grounding` | GUI grounding | 117 | `data/gui_grounding.jsonl` |
| `web_to_html` | Web-to-HTML generation | 100 | `data/web_to_html.jsonl` |
The `visual_search` and `web_to_html` configurations are direct subsets of
existing public datasets, not newly authored evaluation data. Specifically,
all 106 Visual Search examples come from
[VisualProbe_Hard](https://huggingface.co/datasets/Mini-o3/VisualProbe_Hard), and all 100 Web-to-HTML tasks
come from the `Vision2Web-webpage` portion of
[Vision2Web](https://huggingface.co/datasets/zai-org/Vision2Web).
The metadata uses the layout consumed by the accompanying OpenVisTool
evaluation code. Every image path resolves as
`data/images/<source_dataset>/<images[i]>`. Each Web-to-HTML task also includes
its resource directory and `workflow.json` under the same source-dataset root.
## Source composition
| Configuration | Upstream source | Items |
|---|---|---:|
| `chart` | [CharXiv](https://huggingface.co/datasets/princeton-nlp/CharXiv) | 55 |
| `chart` | [ChartMuseum](https://huggingface.co/datasets/lytang/ChartMuseum) | 55 |
| `table` | [MMT-Bench](https://huggingface.co/datasets/Kaining/MMT-Bench) | 112 |
| `table` | [TableVQA-Bench](https://github.com/naver-ai/tablevqabench) | 14 |
| `visual_search` | [VisualProbe_Hard](https://mini-o3.github.io/) | 106 |
| `gui_grounding` | [ScreenSpot-Pro](https://huggingface.co/datasets/likaixin/ScreenSpot-Pro) | 117 |
| `web_to_html` | [Vision2Web-webpage](https://huggingface.co/datasets/zai-org/Vision2Web) | 100 |
The `source_dataset` field preserves the source label for every example.
## Upstream source and license review
| Source label | Canonical source | License |
|---|---|---|
| `CharXiv` | https://huggingface.co/datasets/princeton-nlp/CharXiv | CC BY-SA 4.0 |
| `ChartMuseum` | https://huggingface.co/datasets/lytang/ChartMuseum | CC BY-SA 4.0 |
| `MMTBench` | https://huggingface.co/datasets/Kaining/MMT-Bench | MIT |
| `TableVQA-Bench` | https://github.com/naver-ai/tablevqabench | MIT |
| `VisualProbe_Hard` | https://mini-o3.github.io/ | CC BY-NC 4.0 |
| `ScreenSpot-Pro` | https://huggingface.co/datasets/likaixin/ScreenSpot-Pro | MIT |
| `Vision2Web-webpage` | https://huggingface.co/datasets/zai-org/Vision2Web | CC BY-NC-SA 4.0 |
# OpenVisTool-Bench
OpenVisTool-Bench is a 559-example, five-domain benchmark for visual agents
that can call inspection and editing tools. All five configurations use the
`test` split.
## Configurations
| Configuration | Domain | Items | Metadata |
|---|---|---:|---|
| `chart` | Chart reasoning | 110 | `data/chart.jsonl` |
| `table` | Table reasoning | 126 | `data/table.jsonl` |
| `visual_search` | Fine-grained visual search | 106 | `data/visual_search.jsonl` |
| `gui_grounding` | GUI grounding | 117 | `data/gui_grounding.jsonl` |
| `web_to_html` | Web-to-HTML generation | 100 | `data/web_to_html.jsonl` |
The `visual_search` and `web_to_html` configurations are direct subsets of
existing public datasets, not newly authored evaluation data. Specifically,
all 106 Visual Search examples come from
[VisualProbe_Hard](https://huggingface.co/datasets/Mini-o3/VisualProbe_Hard), and all 100 Web-to-HTML tasks
come from the `Vision2Web-webpage` portion of
[Vision2Web](https://huggingface.co/datasets/zai-org/Vision2Web).
The metadata uses the layout consumed by the accompanying OpenVisTool
evaluation code. Every image path resolves as
`data/images/<source_dataset>/<images[i]>`. Each Web-to-HTML task also includes
its resource directory and `workflow.json` under the same source-dataset root.
## Source composition
| Configuration | Upstream source | Items |
|---|---|---:|
| `chart` | [CharXiv](https://huggingface.co/datasets/princeton-nlp/CharXiv) | 55 |
| `chart` | [ChartMuseum](https://huggingface.co/datasets/lytang/ChartMuseum) | 55 |
| `table` | [MMT-Bench](https://huggingface.co/datasets/Kaining/MMT-Bench) | 112 |
| `table` | [TableVQA-Bench](https://github.com/naver-ai/tablevqabench) | 14 |
| `visual_search` | [VisualProbe_Hard](https://mini-o3.github.io/) | 106 |
| `gui_grounding` | [ScreenSpot-Pro](https://huggingface.co/datasets/likaixin/ScreenSpot-Pro) | 117 |
| `web_to_html` | [Vision2Web-webpage](https://huggingface.co/datasets/zai-org/Vision2Web) | 100 |
The `source_dataset` field preserves the source label for every example.
## Upstream source and license review
| Source label | Canonical source | License |
|---|---|---|
| `CharXiv` | https://huggingface.co/datasets/princeton-nlp/CharXiv | CC BY-SA 4.0 |
| `ChartMuseum` | https://huggingface.co/datasets/lytang/ChartMuseum | CC BY-SA 4.0 |
| `MMTBench` | https://huggingface.co/datasets/Kaining/MMT-Bench | MIT |
| `TableVQA-Bench` | https://github.com/naver-ai/tablevqabench | MIT |
| `VisualProbe_Hard` | https://mini-o3.github.io/ | CC BY-NC 4.0 |
| `ScreenSpot-Pro` | https://huggingface.co/datasets/likaixin/ScreenSpot-Pro | MIT |
| `Vision2Web-webpage` | https://huggingface.co/datasets/zai-org/Vision2Web | CC BY-NC-SA 4.0 | |