UADA3D: Unsupervised Adversarial Domain Adaptation for 3D Object Detection with Sparse LiDAR and Large Domain Gaps

TL;DR

This study introduces Unsupervised Adversarial Domain Adaptation for 3D Object Detection (UADA3D), which uses an adversarial approach to directly learn domain-invariant features in LiDAR-based 3D object detection.

Abstract

In this study, we address a gap in existing unsupervised domain adaptation approaches on LiDAR-based 3D object detection, which have predominantly concentrated on adapting between established, high-density autonomous driving datasets. We focus on sparser point clouds, capturing scenarios from different perspectives: not just from vehicles on the road but also from mobile robots on sidewalks, which encounter significantly different environmental conditions and sensor configurations. We introduce Unsupervised Adversarial Domain Adaptation for 3D Object Detection (UADA3D). UADA3D does not depend on pre-trained source models or teacher-student architectures. Instead, it uses an adversarial approach to directly learn domain-invariant features. We demonstrate its efficacy in various adaptation scenarios, showing significant improvements in both self-driving car and mobile robot domains. Our code is open-source and will be available soon.

Figures (8)

• Figure 1: Comparison of KITTI (source) and robot data (target). We observe that differences in operating environments, sensor positions, and LiDAR density create a large domain gap. This presents a significant challenge for LiDAR-based 3D object detectors, as well as for the task of domain adaptation.
• Figure 2: An overview of UADA3D (black arrows show forwards, and pink backward pass). While the primary task of $f_{\theta_f}$ and $h_{\theta_y}$ is 3D object detection, the discriminator $g_{\theta_D}$ aims to classify the domain of each detected instance. Discriminator's loss, reversed by GRL, encourages the detector to learn features that are not only effective for object detection but also invariant across domains.
• Figure 3: Comparison of objects in LiDAR-CS and robot datasets.
• Figure 4: Average number of points in an object per class.
• Figure 5: Per class $AP_{3D}$ in adaptation experiments Centerpoint. Vehicle, Pedestrian, and Cyclist. UADA3D is our method.