ContextQMD
Back to Transformers

PP-DocLayoutV3

docs/source/en/model_doc/pp_doclayout_v3.md

5.8.05.9 KB
Original Source
<!--Copyright 2026 The HuggingFace Team. All rights reserved. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. ⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be rendered properly in your Markdown viewer. -->

This model was released on 2026-01-29 and added to Hugging Face Transformers on 2026-01-29.

PP-DocLayoutV3

<div class="flex flex-wrap space-x-1"> </div>

Overview

PP-DocLayoutV3 is a unified and high-efficiency model designed for comprehensive layout analysis. It addresses the challenges of complex physical distortions—such as skewing, curving, and adverse lighting—by integrating instance segmentation and reading order prediction into a single, end-to-end framework.

Model Architecture

PP-DocLayoutV3 evolves from a traditional detection-based approach to a robust instance segmentation architecture built upon the RT-DETR framework. Instead of simple bounding boxes, it utilizes a mask-based detection head to predict pixel-accurate segments for layout elements.

Unlike its predecessor, PP-DocLayoutV3 eliminates decoupled stages by embedding a Global Pointer Mechanism directly within the Transformer decoder layers. This allows the model to concurrently output classification labels, precise masks, and logical reading orders in a single forward pass, significantly reducing latency while enhancing parsing precision on complex document layouts.

Usage

Single input inference

The example below demonstrates how to generate text with PP-DocLayoutV3 using [Pipeline] or the [AutoModel].

<hfoptions id="usage"> <hfoption id="Pipeline">
python
import requests
from PIL import Image

from transformers import pipeline


image = Image.open(requests.get("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout_demo.jpg", stream=True).raw)
layout_detector = pipeline("object-detection", model="PaddlePaddle/PP-DocLayoutV3_safetensors")
results = layout_detector(image)
for idx, res in enumerate(results):
    print(f"Order {idx + 1}: {res}")
</hfoption> <hfoption id="AutoModel">
python
import requests
from PIL import Image

from transformers import AutoImageProcessor, AutoModelForObjectDetection


model_path = "PaddlePaddle/PP-DocLayoutV3_safetensors"
model = AutoModelForObjectDetection.from_pretrained(model_path, device_map="auto")
image_processor = AutoImageProcessor.from_pretrained(model_path)

image = Image.open(requests.get("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout_demo.jpg", stream=True).raw)
inputs = image_processor(images=image, return_tensors="pt").to(model.device)

outputs = model(**inputs)
results = image_processor.post_process_object_detection(outputs, target_sizes=[image.size[::-1]])
for result in results:
    for idx, (score, label_id, box, polygon_points) in enumerate(zip(result["scores"], result["labels"], result["boxes"], result["polygon_points"])):
        score, label = score.item(), label_id.item()
        box = [round(i, 2) for i in box.tolist()]
        print(f"Order {idx + 1}: {model.config.id2label[label]}, score: {score:.2f}, box: {box}, polygon_points: {polygon_points}")
</hfoption> </hfoptions>

Batched inference

PP-DocLayoutV3 also supports batched inference. Here is how you can do it with PP-DocLayoutV3 using [Pipeline] or the [AutoModel]:

<hfoptions id="usage"> <hfoption id="Pipeline">
python
import requests
from PIL import Image

from transformers import pipeline


image = Image.open(requests.get("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout_demo.jpg", stream=True).raw)
layout_detector = pipeline("object-detection", model="PaddlePaddle/PP-DocLayoutV3_safetensors")
results = layout_detector([image, image])
for result in results:
    print("result:")
    for idx, res in enumerate(result):
        print(f"Order {idx + 1}: {res}")
</hfoption> <hfoption id="AutoModel">
python
import requests
from PIL import Image

from transformers import AutoImageProcessor, AutoModelForObjectDetection


model_path = "PaddlePaddle/PP-DocLayoutV3_safetensors"
model = AutoModelForObjectDetection.from_pretrained(model_path, device_map="auto")
image_processor = AutoImageProcessor.from_pretrained(model_path)

image = Image.open(requests.get("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout_demo.jpg", stream=True).raw)
inputs = image_processor(images=[image, image], return_tensors="pt").to(model.device)
target_sizes = [image.size[::-1], image.size[::-1]]

outputs = model(**inputs)
results = image_processor.post_process_object_detection(outputs, target_sizes=target_sizes)
for result in results:
    print("result:")
    for idx, (score, label_id, box, polygon_points) in enumerate(zip(result["scores"], result["labels"], result["boxes"], result["polygon_points"])):
        score, label = score.item(), label_id.item()
        box = [round(i, 2) for i in box.tolist()]
        print(f"Order {idx + 1}: {model.config.id2label[label]}, score: {score:.2f}, box: {box}, polygon_points: {polygon_points}")
</hfoption> </hfoptions>

PPDocLayoutV3ForObjectDetection

[[autodoc]] PPDocLayoutV3ForObjectDetection - forward

PPDocLayoutV3Model

[[autodoc]] PPDocLayoutV3Model

PPDocLayoutV3Config

[[autodoc]] PPDocLayoutV3Config

PPDocLayoutV3ImageProcessor

[[autodoc]] PPDocLayoutV3ImageProcessor - preprocess