Mogadishu Street-View Imagery Set
Share this post:
Mogadishu-Vision: Mapping the Unmapped
The Why
Modern systems for autonomous navigation, urban analytics, and smart infrastructure planning depend heavily on high-quality visual datasets.
However, the vast majority of existing computer vision datasets are built around Western urban environments, leaving cities in the Global South largely absent from the data used to train modern AI systems.
This imbalance creates a major limitation: models trained on Western road structures, traffic patterns, and urban layouts often fail to generalize to cities like Mogadishu.
Somali cities present unique visual characteristics—from road infrastructure and traffic dynamics to informal urban structures—that are rarely represented in existing datasets.
Mogadishu-Vision was created to address this gap as the first effort to build a Visual SLAM (Simultaneous Localization and Mapping) dataset specifically designed for Somali urban environments.
The How
Using smartphone-based 4K video capture combined with GPS synchronization, we collected large-scale visual data across more than 20 kilometers of Mogadishu’s core urban infrastructure.
The collection pipeline was designed to be lightweight, scalable, and deployable in low-resource environments, enabling continuous dataset expansion without specialized hardware.
Collected footage was processed and structured into sequences suitable for Visual SLAM research and real-world localization systems.
To make the dataset useful for modern computer vision models, we implemented COCO-format annotations, allowing compatibility with widely used object detection and segmentation frameworks.
The labeling process identifies key elements such as:
local landmarks
road structures and street layouts
traffic patterns
urban obstacles specific to Somali environments
Particular attention was given to capturing visual conditions common in Somali cities, including mixed infrastructure, informal road markings, and diverse street activity.
The Result
Mogadishu-Vision establishes a foundational visual dataset for Somali urban AI systems, enabling research and development in navigation, mapping, and urban analytics.
The dataset is specifically optimized for Edge AI applications, meaning models trained on it can operate directly on mobile devices and embedded systems without requiring continuous internet connectivity.
This design makes the dataset particularly suitable for real-world applications in environments where connectivity may be limited.
Potential applications enabled by Mogadishu-Vision include:
autonomous navigation systems
urban mapping and digital twin infrastructure
traffic and mobility analysis
disaster response and emergency navigation
mobile-based smart city tools
By grounding computer vision models in real Somali urban data, Mogadishu-Vision represents a major step toward building AI systems that can see and understand Somali cities as they truly exist.