Reinforced Axial Refinement Network for Monocular 3D Object Detection

TL;DR

This paper introduces RAR-Net, a reinforcement learning-based post-processing framework that refines monocular 3D object detection predictions efficiently, improving accuracy with minimal additional computational cost.

Contribution

It proposes a novel reinforcement learning approach for iterative refinement of 3D detection, enhancing existing methods without retraining the entire model.

Findings

Improves detection accuracy on KITTI dataset

Requires minimal extra computational resources

Compatible with various existing detection methods

Abstract

Monocular 3D object detection aims to extract the 3D position and properties of objects from a 2D input image. This is an ill-posed problem with a major difficulty lying in the information loss by depth-agnostic cameras. Conventional approaches sample 3D bounding boxes from the space and infer the relationship between the target object and each of them, however, the probability of effective samples is relatively small in the 3D space. To improve the efficiency of sampling, we propose to start with an initial prediction and refine it gradually towards the ground truth, with only one 3d parameter changed in each step. This requires designing a policy which gets a reward after several steps, and thus we adopt reinforcement learning to optimize it. The proposed framework, Reinforced Axial Refinement Network (RAR-Net), serves as a post-processing stage which can be freely integrated into existing monocular 3D detection methods, and improve the performance on the KITTI dataset with small extra computational costs.