Smart Placement Enhanced Vision: Enhancing 3D-Detection With Learned 3D Placement
Keywords: Synthetic data; 3D object detection; Data augmentation
TL;DR: We propose a 3D-aware placement model to realistically place objects in a road scene to generate data augmentation for 3D detection.
Abstract: The diversity and scale of annotated real-world 3D datasets limit the performance of monocular 3D detectors. Although data augmentation holds potential, creating realistic, scene-aware augmentations for outdoor environments presents a significant challenge.
Existing augmentation methods majorly focus on realistic object appearance by advancing the rendering quality. However, we show that object placement is equally important for downstream 3D detection performance. The main challenge, however, for realistic placement, is to automatically identify the plausible physical properties (location, scale, and orientation) for placing objects in real-world scenes. To this end, we propose Smart-Placement, a novel 3D scene-aware augmentation method for generating diverse and realistic augmentations. In particular, given a background scene, we train a placement network to learn a distribution over plausible 3D bounding boxes. Subsequently, we render realistic cars from 3D assets and place them according to the locations sampled from the learned distribution. Through extensive empirical evaluation on standard benchmark datasets - KITTI and NuScenes, we show that our proposed augmentation method significantly boosts the performance of several existing monocular 3D detectors, setting a new state-of-the-art benchmark, while being highly data efficient.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 2166
Loading