• [2026/08/27] RPM-10K dataset can also be downloaded from [Dropbox]

• [2026/04/08] RPM-10K dataset is now publicly available at Baidu Netdisk. [https://pan.baidu.com/s/17TGCwMqBx2KHdcZKgrpR0Q?pwd=r2a6]

• [2026/07/21] Model weights is now publicly available. [https://pan.baidu.com/s/1abzTBOimkcuqmJZrChhJrg?pwd=64md]

• ✅ RPM-10K: large-scale pointer meter reading dataset
• ✅ DialBench: evaluation benchmark for large foundation models
• ✅ Simple and strong multimodal baseline for pointer meter reading

RPM-10K is designed for accurate and robust pointer meter reading.

• Scale: 10,730 images

• Focus: diverse real-world pointer meters

• Download the dataset from [Dropbox]

DialBench provides a comprehensive benchmark for evaluating pointer meter reading in multimodal LLMs / VLMs.

Features:

• Multiple metrics: Acc_ε, Acc_θ, Ref↓, Rel↓\
• Comparison to both open-source and closed-source VLMs

• Our Model Weights (TBD): ()\

conda create -n dialbench python=3.9
conda activate dialbench

git clone https://github.com/Event-AHU/DialBench.git
cd DialBench
pip install -e .

Run:

bash train.sh

• Modify dataset paths in 'caption_builder.py'

  datasets['train'] = dataset_cls(
              vis_processor=self.vis_processors["train"],
              text_processor=self.text_processors["train"],
              ann_paths=[os.path.join(storage_path, '')], 
              vis_root=vis_root,
          )

bash test.sh

• Evaluation settings are also directly controlled via test.sh

If you find DialBench useful:

@article{wang2025dialbench,
  title={DialBench: Towards Accurate Reading Recognition of Pointer Meter using Large Foundation Models},
  author={Wang, Futian and Weng, Chaoliu and Wang, Xiao and Chen, Zhen and Zhao, Zhicheng and Tang, Jin},
  journal={arXiv preprint arXiv:2511.21982},
  year={2025}
}

• BLIVA
• BLIP-2
• LAVIS
• All open-source contributors

• Code: BSD 3-Clause License
• Dataset: TBD
• Model weights: TBD