Adaptive Object Detection Using Adjacency and Zoom Prediction
This improves object detection efficiency for applications with sparse or small objects, though it is incremental as it builds on existing anchor-based methods.
The paper tackles the inefficiency of fixed anchor regions in object detection by proposing an adaptive search strategy that directs computational resources to sub-regions likely to contain objects, achieving comparable accuracy to Faster R-CNN while using two orders of magnitude fewer anchors on average.
State-of-the-art object detection systems rely on an accurate set of region proposals. Several recent methods use a neural network architecture to hypothesize promising object locations. While these approaches are computationally efficient, they rely on fixed image regions as anchors for predictions. In this paper we propose to use a search strategy that adaptively directs computational resources to sub-regions likely to contain objects. Compared to methods based on fixed anchor locations, our approach naturally adapts to cases where object instances are sparse and small. Our approach is comparable in terms of accuracy to the state-of-the-art Faster R-CNN approach while using two orders of magnitude fewer anchors on average. Code is publicly available.