---

🔥 News

• 2026/09/10 - We have renamed NaviDC-OCR to TeleOCR, and all subsequent model iterations will be developed and released under the TeleOCR version.
• 2026/09/01 - We noticed that EMNLP 2026 is hosting the Dr.DocBench Challenge, a document parsing competition. We evaluated NaviDC-OCR with its native weights, achieving better results than MinerU 2.5 Pro and PaddleOCR-VL 1.6. Detailed results are shown below dr.docbench-challenge. We welcome the use of NaviDC‑OCR for competitions. Going forward, we will continue to deliver competitive parsing models for the community.
• 2026/08/29 — Thanks to Nandraj for the GGUF conversion and llama.cpp support 🔗 NaviDC-OCR-GGUF, and to the community for sharing their experience deploying NaviDC-OCR on Ascend 910B!
• 2026/08/17 — NaviDC-OCR model weights and technical report have been released.

---

📖 Introduction

TeleOCR is a lightweight (~1.2B parameters), open-source Vision-Language Model designed specifically for document understanding.

Unlike existing approaches that primarily focus on either digital documents or camera-captured documents, TeleOCR provides a unified framework for both scenarios.

TeleOCR introduces the following key techniques:

• Multi-node Consensus Voting (MCV) for automatic pseudo-label generation
• Geometry-aware Document Modeling for camera-captured documents
• Curvature-Guided Douglas-Peucker Sampling (CGDP)
• Image-to-Image Self-Verification for automatic data refinement
• Progressive Four-Stage Training pipeline
• Content-Structure Decoupled Learning for tables and formulas

These techniques enable TeleOCR to achieve strong performance across both digital and camera-captured document benchmarks while maintaining a lightweight architecture suitable for practical deployment.

---

⚙️ Data Engine

TeleOCR adopts an automated data engine consisting of four major components:

Most of the generated training data requires no manual annotation.

---

🧠 Progressive Training

TeleOCR is trained through a progressive four-stage training pipeline:

This progressive strategy gradually improves the model from basic vision-language alignment to fine-grained document understanding and structured output generation.

---

📊 Experimental Results

TeleOCR is evaluated on OmniDocBench v1.6, Wild-OmniDocBench, PureDocBench, and the ICDAR 2026 Sci-ImageMiner Challenge.

Bold indicates the best result, while underline indicates the second-best result. It should be noted that all metrics reported in this report were evaluated using the Docker image provided by OmniDocBench v1.6. For Wild-OmniDocBench, we used the Docker image provided by OmniDocBench v1.6 for evaluation, with the metrics obtained by inspecting the Sub-v1.5 results. PureDocBench was also evaluated using the Docker image provided by OmniDocBench v1.6; however, the ground-truth JSON files were converted from the files provided by PureDocBench. In addition, for the real-world degraded tracks in Wild-OmniDocBench and PureDocBench, we recommend enabling LAYOUT_MODE="Segmentation".

---

Layout Visualization of Distorted Documents

To evaluate the model's ability to understand complex document deformations, we conduct a visual evaluation on the public dewarping datasets DocUNet and DIR300, with representative results shown in Figure. TeleOCR directly performs layout and content parsing on distorted documents without dewarping preprocessing or a dedicated rectification model, demonstrating robust parsing under complex geometric deformations.

---

Dr.DocBench Challenge

---

OmniDocBench v1.6

OmniDocBench

---

Wild-OmniDocBench

Wild-OmniDocBench

---

PureDocBench

PureDocBench

---

ICDAR 2026 Sci-ImageMiner Challenge

Challenge Website

---

🚀 Installation & Usage

1. Environment Installation

TeleOCR requires Python 3.10+ and a CUDA-enabled environment for GPU inference.

git clone https://github.com/caipeng328/TeleOCR.git
cd TeleOCR
conda create -n teleocr python=3.10 -y
conda activate teleocr

Install Dependencies

Install TeleOCR and its dependencies in editable mode:

pip install -e .

The -e option installs the project in editable mode, allowing modifications to the source code to take effect immediately without reinstalling the package.

---

2. Inference

TeleOCR provides infer.py for batch inference on document images.

Before running inference, configure:

• Input image directory
• Output directory
• Model path

A typical inference command is:

#!/bin/bash
set -e

IMAGE_SUB_PATH="/path/to/input/images"
RESULT_SAVE_PATH="/path/to/output/results"

python infer.py \
    --image_sub_path "${IMAGE_SUB_PATH}" \
    --result_save_path "${RESULT_SAVE_PATH}" \
    --use_async \
    --override \
        model_path="StarDoc-AI/TeleOCR" \
        BACKEND="vllm-async-engine" \
        LAYOUT_MODE="Detection"

Main Configuration Parameters

---

📝 Citation

If you find TeleOCR useful in your research, please consider citing:

@article{teleocr,
  title={TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents},
  author={Cai, Peng and Zou, Zhaofan and Liu, Shifa and Wang, Yikun and Tang, Jiawei and Yang, Kaicheng and Tong, Meng and He, Zhongjiang and Sun, Hao},
  journal={arXiv preprint arXiv:2608.12898},
  year={2026}
}

---

🙏 Acknowledgements

TeleOCR is built upon the following excellent open-source projects:

• MinerU
• Qwen2.5-VL
• Qwen3
• Transformers
• PyTorch
• FlashAttention

We sincerely thank the authors and contributors of these projects for their valuable contributions to the open-source community.

---

📮 Contact

If you have any questions, suggestions, or issues, please feel free to:

• Open an issue in this repository
• Contact the TeleOCR authors