Weakly Supervised Object Detection

Weakly supervised object detection (WSOD) aims to train object detectors using only image-level labels, significantly reducing annotation costs compared to fully supervised methods. Current research focuses on improving the accuracy of WSOD by leveraging techniques like contrastive learning, self-training, and incorporating additional information such as hallucinated motion or depth, often within transformer-based or multiple instance learning frameworks. These advancements hold significant promise for applications requiring large-scale object detection where obtaining precise bounding box annotations is impractical, such as medical image analysis and remote sensing.