R(Det)^2: Randomized Decision Routing for Object Detection
This work addresses a specific bottleneck in object detection for computer vision researchers, offering an incremental improvement over existing detectors.
The paper tackles the problem of designing high-performance decision heads for object detection by proposing R(Det)^2, which combines decision trees and deep neural networks with randomized decision routing, achieving 1.4-3.6% AP improvement on MS-COCO.
In the paradigm of object detection, the decision head is an important part, which affects detection performance significantly. Yet how to design a high-performance decision head remains to be an open issue. In this paper, we propose a novel approach to combine decision trees and deep neural networks in an end-to-end learning manner for object detection. First, we disentangle the decision choices and prediction values by plugging soft decision trees into neural networks. To facilitate effective learning, we propose randomized decision routing with node selective and associative losses, which can boost the feature representative learning and network decision simultaneously. Second, we develop the decision head for object detection with narrow branches to generate the routing probabilities and masks, for the purpose of obtaining divergent decisions from different nodes. We name this approach as the randomized decision routing for object detection, abbreviated as R(Det)$^2$. Experiments on MS-COCO dataset demonstrate that R(Det)$^2$ is effective to improve the detection performance. Equipped with existing detectors, it achieves $1.4\sim 3.6$\% AP improvement.