This is the model weights for PP-DocLayoutv3 in safetensors format. Get PaddlePaddle weights at
PP-DocLayoutV31import requests
2from PIL import Image
3from transformers import AutoImageProcessor, AutoModelForObjectDetection
4
5model_path = "PaddlePaddle/PP-DocLayoutV3_safetensors"
6model = AutoModelForObjectDetection.from_pretrained(model_path)
7image_processor = AutoImageProcessor.from_pretrained(model_path)
8
9image = Image.open(requests.get("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout_demo.jpg", stream=True).raw)
10inputs = image_processor(images=image, return_tensors="pt")
11
12outputs = model(**inputs)
13results = image_processor.post_process_object_detection(outputs, target_sizes=[image.size[::-1]])
14for result in results:
15 for idx, (score, label_id, box, polygon_points) in enumerate(zip(result["scores"], result["labels"], result["boxes"], result["polygon_points"])):
16 score, label = score.item(), label_id.item()
17 box = [round(i, 2) for i in box.tolist()]
18 print(f"Order {idx + 1}: {model.config.id2label[label]}, score: {score:.2f}, box: {box}, polygon_points: {polygon_points}")
If you find PP-DocLayoutV3 helpful, feel free to give us a star and citation.
1@misc{cui2026rtdoclayoutrealtimeendtoenddocument,
2 title={RT-DocLayout: Real-Time End-to-End Document Layout Analysis with Reading Order in the Wild},
3 author={Cheng Cui and Tingquan Gao and Xueqing Wang and Changda Zhou and Hongen Liu and Ting Sun and Yubo Zhang and Zelun Zhang and Jiaxuan Liu and Manhui Lin and Yue Zhang and Suyin Liang and Yiqing Xiang and Yi Liu},
4 year={2026},
5 eprint={2606.23344},
6 archivePrefix={arXiv},
7 primaryClass={cs.CV},
8 url={https://arxiv.org/abs/2606.23344},
9}
10}
11
