On Hyperbolic Embeddings in 2D Object Detection
This work addresses the problem of improving object detection accuracy by exploring hyperbolic embeddings, offering a novel approach for computer vision researchers and practitioners.
The authors investigated whether hyperbolic geometry better matches the underlying structure of object classification space in 2D object detection, finding that it leads to lower classification errors and boosts overall performance on large-scale, long-tailed, and zero-shot benchmarks.
Object detection, for the most part, has been formulated in the euclidean space, where euclidean or spherical geodesic distances measure the similarity of an image region to an object class prototype. In this work, we study whether a hyperbolic geometry better matches the underlying structure of the object classification space. We incorporate a hyperbolic classifier in two-stage, keypoint-based, and transformer-based object detection architectures and evaluate them on large-scale, long-tailed, and zero-shot object detection benchmarks. In our extensive experimental evaluations, we observe categorical class hierarchies emerging in the structure of the classification space, resulting in lower classification errors and boosting the overall object detection performance.