COD

COD detects concealed objects embedded in complex backgrounds to enable accurate detection and segmentation of objects that exhibit high intrinsic similarity to their surroundings.

Key Features:

• Large-Scale Dataset - COD10K: A dataset of 10,000 images covering 78 object categories for training and evaluation of concealed object detection models.
• Rich Annotations: Per-image annotations including object categories, object boundaries, challenging attributes, object-level labels, and instance-level annotations.
• Search Identification Network (SINet): A baseline architecture inspired by natural hunting strategies that demonstrates superior performance compared to twelve advanced baselines across multiple datasets.

Scientific Applications:

• Surveillance Systems: Identifying hidden threats or contraband that blend into complex backgrounds.
• Medical Imaging: Detecting anomalies concealed within medical scans to support diagnostic analysis.
• Robotics and Autonomous Vehicles: Improving environmental understanding and obstacle detection in visually cluttered scenes.

Methodology:

Systematic collection and annotation of the COD10K dataset; development of SINet using strategies inspired by natural hunting behaviors; and comprehensive evaluation against existing baselines, including comparison to twelve advanced methods across datasets.