Datasets:
Tasks:
Question Answering
Modalities:
Image
Formats:
imagefolder
Languages:
English
Size:
1K - 10K
ArXiv:
Tags:
math
License:
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
task_categories:
|
4 |
+
- question-answering
|
5 |
+
language:
|
6 |
+
- en
|
7 |
+
tags:
|
8 |
+
- math
|
9 |
+
---
|
10 |
+
|
11 |
+
## Overview
|
12 |
+
VCBench provides a standardized framework for evaluating vision-language models. This document outlines the procedures for both standard evaluation and GPT-assisted evaluation of your model's outputs.
|
13 |
+
|
14 |
+
## 1. Standard Evaluation
|
15 |
+
|
16 |
+
### 1.1 Output Format Requirements
|
17 |
+
Models must produce outputs in JSONL format with the following structure:
|
18 |
+
```
|
19 |
+
{"id": <int>, "pred_answer": "<answer_letter>"}
|
20 |
+
{"id": <int>, "pred_answer": "<answer_letter>"}
|
21 |
+
...
|
22 |
+
```
|
23 |
+
|
24 |
+
**Example File (`submit.jsonl`):**
|
25 |
+
```json
|
26 |
+
{"id": 1, "pred_answer": "A"}
|
27 |
+
{"id": 2, "pred_answer": "B"}
|
28 |
+
{"id": 3, "pred_answer": "C"}
|
29 |
+
```
|
30 |
+
|
31 |
+
### 1.2 Evaluation Procedure
|
32 |
+
1. Ensure your predictions file follows the specified format
|
33 |
+
2. Run the evaluation script:
|
34 |
+
```bash
|
35 |
+
python evaluate_vcbench.py -p ./path/to/predictions.jsonl -g ./path/to/VCBench_with_answer.json
|
36 |
+
```
|
37 |
+
`VCBench_with_answer.json` is the ground truth file which can be downloaded from [here](https://huggingface.co/datasets/cloudcatcher2/VCBench/resolve/main/VCBench_with_answer.json).
|
38 |
+
|
39 |
+
## 2. GPT-Assisted Evaluation
|
40 |
+
|
41 |
+
### 2.1 Output Format Requirements
|
42 |
+
For natural language responses, use this JSONL format:
|
43 |
+
```
|
44 |
+
{"id": <int>, "pred_answer": "<natural_language_response>"}
|
45 |
+
{"id": <int>, "pred_answer": "<natural_language_response>"}
|
46 |
+
...
|
47 |
+
```
|
48 |
+
|
49 |
+
**Example File (`nl_predictions.jsonl`):**
|
50 |
+
```json
|
51 |
+
{"id": 1, "pred_answer": "The correct answer is A"}
|
52 |
+
{"id": 2, "pred_answer": "After careful analysis, option B appears correct"}
|
53 |
+
{"id": 3, "pred_answer": "C is the right choice"}
|
54 |
+
```
|
55 |
+
|
56 |
+
### 2.2 Environment Setup
|
57 |
+
Set your Dashscope API key:
|
58 |
+
```bash
|
59 |
+
export DASHSCOPE_KEY="your_api_key_here"
|
60 |
+
```
|
61 |
+
|
62 |
+
### 2.3 Evaluation Procedure
|
63 |
+
```bash
|
64 |
+
python evaluate_vcbench_by_gpt.py -p ./path/to/nl_predictions.jsonl -g ./path/to/VCBench_with_answer.json
|
65 |
+
```
|
66 |
+
|
67 |
+
## 3. Expected Output
|
68 |
+
Both evaluation scripts will provide:
|
69 |
+
- Overall accuracy percentage
|
70 |
+
- Per-question-type accuracy breakdown
|
71 |
+
- Progress updates during evaluation
|
72 |
+
|
73 |
+
|
74 |
+
|
75 |
+
## Citation
|
76 |
+
|
77 |
+
<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->
|
78 |
+
|
79 |
+
**BibTeX:**
|
80 |
+
|
81 |
+
```bibtex
|
82 |
+
@misc{wong2025vcbench
|
83 |
+
author = {Zhikai Wang and Jiashuo Sun and Wenqi Zhang and Zhiqiang Hu and Xin Li and Fan Wang and Deli Zhao},
|
84 |
+
title = {Benchmarking Multimodal Mathematical Reasoning with Explicit Visual Dependency},
|
85 |
+
year = {2025},
|
86 |
+
eprint = {2504.18589},
|
87 |
+
archivePrefix = {arxiv},
|
88 |
+
primaryClass = {cs.CV},
|
89 |
+
url = {https://arxiv.org/abs/2504.18589}
|
90 |
+
}
|
91 |
+
```
|
92 |
+
|
93 |
+
## Dataset Card Authors
|
94 |
+
|
95 |
+
- [Zhikai Wang](https://cloudcatcher888.github.io/): [email protected]
|
96 |
+
- [Jiashuo Sun](https://gasolsun36.github.io/): [email protected]
|