PP-DocBee2: Improved Baselines with Efficient Data for Multimodal Document Understanding
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866918069624897536 |
|---|---|
| author | Huang, Kui Chen, Xinrong Lv, Wenyu Liao, Jincheng Wang, Guanzhong Liu, Yi |
| author_facet | Huang, Kui Chen, Xinrong Lv, Wenyu Liao, Jincheng Wang, Guanzhong Liu, Yi |
| contents | This report introduces PP-DocBee2, an advanced version of the PP-DocBee, designed to enhance multimodal document understanding. Built on a large multimodal model architecture, PP-DocBee2 addresses the limitations of its predecessor through key technological improvements, including enhanced synthetic data quality, improved visual feature fusion strategy, and optimized inference methodologies. These enhancements yield an $11.4\%$ performance boost on internal benchmarks for Chinese business documents, and reduce inference latency by $73.0\%$ to the vanilla version. A key innovation of our work is a data quality optimization strategy for multimodal document tasks. By employing a large-scale multimodal pre-trained model to evaluate data, we apply a novel statistical criterion to filter outliers, ensuring high-quality training data. Inspired by insights into underutilized intermediate features in multimodal models, we enhance the ViT representational capacity by decomposing it into layers and applying a novel feature fusion strategy to improve complex reasoning. The source code and pre-trained model are available at \href{https://github.com/PaddlePaddle/PaddleMIX}{https://github.com/PaddlePaddle/PaddleMIX}. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_18023 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | PP-DocBee2: Improved Baselines with Efficient Data for Multimodal Document Understanding Huang, Kui Chen, Xinrong Lv, Wenyu Liao, Jincheng Wang, Guanzhong Liu, Yi Computer Vision and Pattern Recognition Artificial Intelligence Computation and Language This report introduces PP-DocBee2, an advanced version of the PP-DocBee, designed to enhance multimodal document understanding. Built on a large multimodal model architecture, PP-DocBee2 addresses the limitations of its predecessor through key technological improvements, including enhanced synthetic data quality, improved visual feature fusion strategy, and optimized inference methodologies. These enhancements yield an $11.4\%$ performance boost on internal benchmarks for Chinese business documents, and reduce inference latency by $73.0\%$ to the vanilla version. A key innovation of our work is a data quality optimization strategy for multimodal document tasks. By employing a large-scale multimodal pre-trained model to evaluate data, we apply a novel statistical criterion to filter outliers, ensuring high-quality training data. Inspired by insights into underutilized intermediate features in multimodal models, we enhance the ViT representational capacity by decomposing it into layers and applying a novel feature fusion strategy to improve complex reasoning. The source code and pre-trained model are available at \href{https://github.com/PaddlePaddle/PaddleMIX}{https://github.com/PaddlePaddle/PaddleMIX}. |
| title | PP-DocBee2: Improved Baselines with Efficient Data for Multimodal Document Understanding |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2506.18023 |