asmaatagelsir/Sudan-MM-2025-Metadata-Architecture

Multimodal dataset architecture for Sudanese culture and dialect, featuring 695 image/video pairs with MSA and Sudanese Arabic captions. Developed by Team 4sparks for the Sudan-MM 2025 initiative # Sudan-MM-2025-Metadata-Architecture Multimodal dataset architecture for Sudanese culture and dialect, featuring 695 image/video pairs with MSA and Sudanese Arabic captions. Developed by Team 4sparks for the Sudan-MM 2025 initiative # Sudan-MM 2025: Multimodal Dataset Architecture This project represents the technical documentation and metadata architecture for the **Sudan-MM 2025** dataset, developed by **Team 4sparks**. ## Project Overview Sudan-MM 2025 is a multimodal dataset designed to bridge the linguistic gap in African AI. It features 695 pairs (640 images and 55 videos) representing daily life, culture, and environments across various states in Sudan. ## My Role (Asmaa Tagelsir) As a core member of Team 4sparks, my primary responsibilities included: - **Media Collection:** Sourcing 700+ high-quality images and videos ensuring geographic diversity across Sudan. - **Linguistic Annotation:** Writing and refining authentic captions in both **Modern Standard Arabic (MSA)** and **Sudanese Arabic dialect**. - **Ethical Sourcing:** Managing permissions and licensing from local content creators to ensure a legally sound and ethically compliant dataset. ## Technical Specifications - **Format:** Image/Video (.jpg, .mp4) with synchronized Audio (.mp3). - **Categories:** Agriculture, Public Infrastructure, Marketplaces, and Local Culture. - **Goal:** To support the development of Sovereign AI that understands Sudanese linguistic nuances. ## Team: 4sparks - Nafisa Ahmed - Rafa Abdallah - Miada Abdalmonim - Asmaa Tagelsir ## Dataset Access The multimodal media files (640 images, 55 videos, and 695 voice recordings) are hosted on Google Drive for high-resolution integrity. - **Dataset Root Folder:** [drive.google.com] - **Metadata Inventory:** The `metadata.csv` file in this repository contains the synchronized captions and file IDs.