|
|
|
|
|
--- |
|
|
--- |
|
|
library_name: ultralytics |
|
|
tags: |
|
|
- document-layout |
|
|
- multi-lingual |
|
|
- yolo |
|
|
- layout-analysis |
|
|
- allow_download_tracking |
|
|
- doclayout-yolo |
|
|
license: mit |
|
|
datasets: |
|
|
- indicdlp |
|
|
pipeline_tag: object-detection |
|
|
--- |
|
|
--- |
|
|
|
|
|
# 📄 IndicDLP: |
|
|
# A Foundational Dataset for Multi-Lingual and Multi-Domain Document Layout Parsing |
|
|
|
|
|
|
|
|
**[ICDAR 2025 (Oral) — 🏆 Best Student Paper Runner-Up Award](https://www.icdar2025.com/)** |
|
|
|
|
|
[](https://link_to_paper.com) |
|
|
[](https://indicdlp.github.io/) |
|
|
[](https://github.com/Indic-Layout/IndicDLP) |
|
|
|
|
|
|
|
|
#### 👩💻 Authors: |
|
|
- Oikantik Nath (IIT Madras) |
|
|
- Sahithi Kukkala (IIIT Hyderabad) |
|
|
- Mitesh Khapra (IIT Madras) |
|
|
- Ravi Kiran Sarvadevabhatla (IIIT Hyderabad) |
|
|
|
|
|
--- |
|
|
|
|
|
|
|
|
## IndicDLP Dataset |
|
|
|
|
|
IndicDLP is a large-scale, foundational dataset created to advance document layout parsing in multi-lingual and multi-domain settings. It comprises 119,806 document images covering 11 Indic languages and English: Assamese, Bengali, English, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, and Telugu. The dataset spans 12 diverse document categories, including Novels, Textbooks, Magazines, Acts & Rules, Research Papers, Manuals, Brochures, Syllabi, Question Papers, Notices, Forms, and Newspapers. |
|
|
|
|
|
The dataset contains 42 physical and logical layout classes. IndicDLP includes both digitally-born and scanned documents, with annotations created using Shoonya, an open-source tool built on Label Studio. The dataset is curated to support robust layout understanding across diverse scripts, domains, and document types. |
|
|
|
|
|
|
|
|
## IndicDLP Model Checkpoints |
|
|
|
|
|
We provide 3 model checkpoints — YOLOv10x, DocLayout-YOLO, and RoDLA — finetuned on the IndicDLP dataset. These models are optimized for robust document layout parsing across a wide range of Indic languages and document types, and are capable of detecting all 42 region labels defined in the dataset. |
|
|
|
|
|
These checkpoints have demonstrated strong performance on both scanned and digitally-born documents. They are ready to use for inference, serve as strong baselines for benchmarking, and can be further fine-tuned for downstream tasks such as structure extraction or semantic tagging. |
|
|
|
|
|
|
|
|
## 🏆 Available Checkpoints |
|
|
|
|
|
| Model | mAP<sub>[50:95]</sub> | Download File | Framework | |
|
|
|------------------|----------------------|-------------------------------------------|---------------------| |
|
|
| **YOLOv10x** | **57.7** | [yolov10x.pt](./md2_PT_indicdlp_FT.pt) | Ultralytics YOLOv10 | |
|
|
| **DocLayout-YOLO**| 54.5 | [doclayout_yolo.pt](.doclayout_yolo_indicdlp.pt) | DocLayout-YOLO | |
|
|
| **RoDLA** | 53.1 | [rodla.pth](./rodla_internimage_xl_indicdlp.pth) | RoDLA | |
|
|
|
|
|
--- |
|
|
|
|
|
## 🚀 Quick Start |
|
|
|
|
|
Download the desired checkpoint(s) from this page or using the `huggingface_hub` CLI. See example commands for each model below. |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚙️ YOLOv10 |
|
|
|
|
|
### Environment Setup |
|
|
|
|
|
```bash |
|
|
conda create -n indicdlp python=3.12 |
|
|
conda activate indicdlp |
|
|
pip install -r requirements.txt |
|
|
``` |
|
|
|
|
|
### Training |
|
|
|
|
|
```bash |
|
|
yolo detect train \ |
|
|
data=dataset_root/data.yaml \ |
|
|
model=yolov10x.yaml \ |
|
|
device=0,1,2,3,4,5,6,7 \ |
|
|
epochs=100 \ |
|
|
imgsz=1024 \ |
|
|
batch=64 \ |
|
|
name=indicdlp_yolov10x \ |
|
|
patience=5 |
|
|
``` |
|
|
|
|
|
### Evaluation |
|
|
|
|
|
```bash |
|
|
yolo detect val \ |
|
|
model=/path/to/model_weights.pt \ |
|
|
data=dataset_root/data.yaml \ |
|
|
split=test |
|
|
``` |
|
|
|
|
|
### Inference |
|
|
|
|
|
```bash |
|
|
yolo detect predict \ |
|
|
model=/path/to/model_weights.pt \ |
|
|
source=dataset_root/images/test/ \ |
|
|
conf=0.2 \ |
|
|
save=True |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚙️ DocLayout-YOLO |
|
|
|
|
|
For **DocLayout-YOLO** setup, training, and inference instructions, please see [IndicDLP GitHub repository](https://github.com/Indic-Layout/IndicDLP). |
|
|
|
|
|
|
|
|
--- |
|
|
|
|
|
### ⚙️ RoDLA |
|
|
|
|
|
For **RoDLA** setup, training, and inference instructions, please see [IndicDLP GitHub repository](https://github.com/Indic-Layout/IndicDLP). |
|
|
|
|
|
--- |
|
|
|
|
|
## 📦 Dataset |
|
|
|
|
|
These models are finetuned on the [IndicDLP dataset](https://aikosh.indiaai.gov.in/home/datasets/details/indicdlp.html). |
|
|
For details, annotation schema, and scripts, visit the [IndicDLP project homepage](https://indicdlp.github.io/). |
|
|
|
|
|
--- |
|
|
|
|
|
## 📑 Citation |
|
|
|
|
|
If you use these models or the dataset, please cite: |
|
|
|
|
|
```bibtex |
|
|
@InProceedings{10.1007/978-3-032-04614-7_2, |
|
|
author="Nath, Oikantik |
|
|
and Kukkala, Sahithi |
|
|
and Khapra, Mitesh |
|
|
and Sarvadevabhatla, Ravi Kiran", |
|
|
editor="Yin, Xu-Cheng |
|
|
and Karatzas, Dimosthenis |
|
|
and Lopresti, Daniel", |
|
|
title="IndicDLP: A Foundational Dataset for Multi-lingual and Multi-domain Document Layout Parsing", |
|
|
booktitle="Document Analysis and Recognition -- ICDAR 2025", |
|
|
year="2026", |
|
|
publisher="Springer Nature Switzerland", |
|
|
address="Cham", |
|
|
pages="23--39", |
|
|
abstract="Document layout analysis is essential for downstream tasks such as information retrieval, extraction, OCR, and digitization. However, existing large-scale datasets like PubLayNet and DocBank lack fine-grained region labels and multilingual diversity, making them insufficient for representing complex document layouts. Human-annotated datasets such as {\$}{\$}M^{\{}6{\}}Doc{\$}{\$}M6Doc and {\$}{\$}{\backslash}text {\{}D{\}}^{\{}4{\}}{\backslash}text {\{}LA{\}}{\$}{\$}D4LA offer richer labels and greater domain diversity, but are too small to train robust models and lack adequate multilingual coverage. This gap is especially pronounced for Indic documents, which encompass diverse scripts yet remain underrepresented in current datasets, further limiting progress in this space. To address these shortcomings, we introduce IndicDLP, a large-scale foundational document layout dataset spanning 11 representative Indic languages alongside English and 12 common document domains. Additionally, we curate UED-mini, a dataset derived from DocLayNet and {\$}{\$}M^{\{}6{\}}Doc{\$}{\$}M6Doc, to enhance pretraining and provide a solid foundation for Indic layout models. Our experiments demonstrate that fine-tuning existing English models on IndicDLP significantly boosts performance, validating its effectiveness. Moreover, models trained on IndicDLP generalize well beyond Indic layouts, making it a valuable resource for document digitization. This work bridges gaps in scale, diversity, and annotation granularity, driving inclusive and efficient document understanding.", |
|
|
isbn="978-3-032-04614-7" |
|
|
} |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## 🙏 Acknowledgements |
|
|
|
|
|
- [Ultralytics YOLOv10](https://github.com/ultralytics/ultralytics) |
|
|
- [DocLayout-YOLO](https://github.com/opendatalab/DocLayout-YOLOR) |
|
|
- [RoDLA](https://github.com/yufanchen96/RoDLA) |
|
|
|
|
|
--- |
|
|
|
|
|
## 📬 Contact |
|
|
For issues in running code/links not working, please reach out to [Sahithi Kukkala](mailto:sahithi.kukkala@research.iiit.ac.in) or [Oikantik Nath](mailto:oikantik@cse.iitm.ac.in) or mention in the **ISSUES** section. |
|
|
|
|
|
For questions or collaborations, please reach out to [Dr. Ravi Kiran Sarvadevabhatla](mailto:ravi.kiran@iiit.ac.in). |
|
|
|
|
|
--- |
|
|
|
