BD-News-Photocard-Logos: A Dataset of Bangladeshi News Photocard Logos for Forensic Verification (40 Outlets)

Sadat Islam Rahin, Md. Shehab Uddin, Md. Mubtasim Abedin Jaowad, Abu Sayed Md. Mostafizur Rahaman

This dataset contains manually annotated bounding boxes around the logos of 40 Bangladeshi news outlets as they appear on social media news photocards — square image-based news summaries distributed primarily on Facebook. It was created to support the development and evaluation of a multi-layer forensic pipeline for news photocard verification. The dataset comprises 800 original photocards collected from the official Facebook pages of each outlet, annotated with logo-only bounding boxes in YOLOv8 format. After augmentation in Roboflow (horizontal flip, brightness ±15%, blur ≤1px, 3x multiplier), the dataset contains 1,915 images split 70/20/10 into train/valid/test. The 40 outlets span both broadcast television and print media, covering major Bengali-language and English-language publications. A YOLOv8n model trained on Version 2 of this dataset achieved a test-set mAP@50 of 0.9936 across 40 classes. This dataset was produced as part of the undergraduate thesis:"Multi-Layer Forensic Verification of Bangladeshi News Photocards Using Deep Learning and OCR-Based Content Cross-Referencing"Bangladesh University of Professionals, 2026. Note: Images were collected from publicly accessible Facebook pages for non-commercial academic research purposes. Images remain the intellectual property of their respective news organizations.