bengaliAI/badlad-yolov8m-seg

BaDLAD YOLOv8m-seg

YOLOv8 medium segmentation model trained on BaDLAD for Bengali document layout analysis. Used as the layout stage in Bengali.AI OCR pipelines (bbocr / chitrolipi) and for large-scale pseudolabeling.

Paper: BaDLAD (ICDAR 2023)

Code: BengaliAI/BADLAD

Dataset: BaDLAD on Kaggle

Model Details

Field	Value
Architecture	YOLOv8m-seg
Framework	Ultralytics 8.0.59
Classes	paragraph, text_box, image, table (4)
Training data	BaDLAD labeled train
Epochs	20
Batch	32
Image size	640
Optimizer	SGD (lr0=0.01)
Checkpoint	best.pt (~52 MB)
sha256	890e6661af89d92c30f175cc2137e0f677eb425172720b0c77d43454fa09dcce
Trained	2023-04-03

Hyperparameters above are taken from the checkpoint’s embedded train_args (Ultralytics stores the full training config inside every .pt).

Usage

from ultralytics import YOLO
from huggingface_hub import hf_hub_download

weights = hf_hub_download("bengaliAI/badlad-yolov8m-seg", "best.pt")
model = YOLO(weights)
results = model.predict("page.png", conf=0.05, imgsz=640)

Related models

Repo	Role
bengaliAI/badlad-mrcnn-paper	Detectron Mask R-CNN paper baseline
bengaliAI/badlad-frcnn-paper	Detectron Faster R-CNN paper baseline

Citation

@inproceedings{shihab2023badlad,
    title     = {{BaDLAD}: A Large Multi-Domain {Bengali} Document Layout Analysis Dataset},
    author    = {Shihab, Md. Istiak Hossain and Hasan, Md. Rakibul and Emon, Mahfuzur Rahman and Hossen, Syed Mobassir and Ansary, Md. Nazmuddoha and Ahmed, Intesur and Rakib, Fazle Rabbi and Dhruvo, Shahriar Elahi and Dip, Souhardya Saha and Pavel, Akib Hasan and Meghla, Marsia Haque and Haque, Md. Rezwanul and Chowdhury, Sayma Sultana and Sadeque, Farig and Reasat, Tahsin and Humayun, Ahmed Imtiaz and Sushmit, Asif Shahriyar},
    booktitle = {Proceedings of the 17th International Conference on Document Analysis and Recognition (ICDAR)},
    year      = {2023},
    url       = {https://arxiv.org/abs/2303.05325},
}