PM4Bench-Release commited on
Commit
bb381b8
·
1 Parent(s): 3c74902

Pin paper references to arXiv v3

Browse files
Files changed (1) hide show
  1. README.md +6 -6
README.md CHANGED
@@ -33,7 +33,7 @@ configs:
33
 
34
  **Synthetic multilingual OCR data for QGO reinforcement learning**
35
 
36
- [![arXiv](https://img.shields.io/badge/arXiv-2503.18484v3-B31B1B.svg?logo=arxiv&logoColor=white)](https://arxiv.org/html/2503.18484v3)
37
  [![Project Page](https://img.shields.io/badge/Project-Page-145C73.svg?logo=githubpages&logoColor=white)](https://little-bird-vodka.github.io/PM4Bench/)
38
  [![Code](https://img.shields.io/badge/GitHub-Code-181717.svg?logo=github&logoColor=white)](https://github.com/opendatalab/PM4Bench)
39
  [![Benchmark](https://img.shields.io/badge/Hugging_Face-Benchmark-FFD21E.svg)](https://huggingface.co/datasets/songjhPKU/PM4Bench)
@@ -44,7 +44,7 @@ configs:
44
  ## Overview
45
 
46
  The paper [*Benchmarking and Boosting Multilingual Capabilities of LVLMs via
47
- OCR-Centric Reinforcement Learning*](https://arxiv.org/html/2503.18484v3)
48
  uses PM4Bench to show that OCR is a key source of cross-lingual performance
49
  gaps when text is rendered visually. QGO addresses that finding with GRPO on
50
  synthetic OCR data, without task-specific VQA or GUI supervision. This
@@ -76,14 +76,14 @@ includes the launcher and exact reward implementation.
76
 
77
  ## Cite Us
78
 
79
- ```bibtex
80
  @misc{gao2026benchmarkingboostingmultilingualcapabilities,
81
  title={Benchmarking and Boosting Multilingual Capabilities of LVLMs via OCR-Centric Reinforcement Learning},
82
  author={Junyuan Gao and Jiahe Song and Jiang Wu and Runchuan Zhu and Guanlin Shen and Shasha Wang and Xingjian Wei and Haote Yang and Weijia Li and Bin Wang and Lijun Wu and Conghui He},
83
  year={2026},
84
- eprint={2503.18484},
85
  archivePrefix={arXiv},
86
  primaryClass={cs.CV},
87
- url={https://arxiv.org/html/2503.18484v3},
88
  }
89
- ```
 
33
 
34
  **Synthetic multilingual OCR data for QGO reinforcement learning**
35
 
36
+ [![arXiv v3](https://img.shields.io/badge/arXiv-v3-B31B1B.svg?logo=arxiv&logoColor=white)](https://little-bird-vodka.github.io/PM4Bench/paper-v3.html)
37
  [![Project Page](https://img.shields.io/badge/Project-Page-145C73.svg?logo=githubpages&logoColor=white)](https://little-bird-vodka.github.io/PM4Bench/)
38
  [![Code](https://img.shields.io/badge/GitHub-Code-181717.svg?logo=github&logoColor=white)](https://github.com/opendatalab/PM4Bench)
39
  [![Benchmark](https://img.shields.io/badge/Hugging_Face-Benchmark-FFD21E.svg)](https://huggingface.co/datasets/songjhPKU/PM4Bench)
 
44
  ## Overview
45
 
46
  The paper [*Benchmarking and Boosting Multilingual Capabilities of LVLMs via
47
+ OCR-Centric Reinforcement Learning*](https://little-bird-vodka.github.io/PM4Bench/paper-v3.html)
48
  uses PM4Bench to show that OCR is a key source of cross-lingual performance
49
  gaps when text is rendered visually. QGO addresses that finding with GRPO on
50
  synthetic OCR data, without task-specific VQA or GUI supervision. This
 
76
 
77
  ## Cite Us
78
 
79
+ <pre><code class="language-bibtex">
80
  @misc{gao2026benchmarkingboostingmultilingualcapabilities,
81
  title={Benchmarking and Boosting Multilingual Capabilities of LVLMs via OCR-Centric Reinforcement Learning},
82
  author={Junyuan Gao and Jiahe Song and Jiang Wu and Runchuan Zhu and Guanlin Shen and Shasha Wang and Xingjian Wei and Haote Yang and Weijia Li and Bin Wang and Lijun Wu and Conghui He},
83
  year={2026},
84
+ eprint={2503&#46;18484},
85
  archivePrefix={arXiv},
86
  primaryClass={cs.CV},
87
+ url={https://arxiv&#46;org/abs/2503&#46;18484v3},
88
  }
89
+ </code></pre>