SF-UDA$^{3D}$: Source-Free Unsupervised Domain Adaptation for LiDAR-Based 3D Object Detection
| Autor: | Nicu Sebe, Elisa Ricci, Fabio Galasso, Cristiano Saltori, Stéphane Lathuilière |
|---|---|
| Přispěvatelé: | University of Trento [Trento], Multimédia (MM), Laboratoire Traitement et Communication de l'Information (LTCI), Institut Mines-Télécom [Paris] (IMT)-Télécom Paris-Institut Mines-Télécom [Paris] (IMT)-Télécom Paris, Département Images, Données, Signal (IDS), Télécom ParisTech, Università degli Studi di Roma 'La Sapienza' = Sapienza University [Rome] |
| Jazyk: | angličtina |
| Rok vydání: | 2020 |
| Předmět: |
FOS: Computer and information sciences
LiDar Computer science domain adaptation Computer Vision and Pattern Recognition (cs.CV) Computer Science - Computer Vision and Pattern Recognition ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Point cloud 02 engineering and technology 010501 environmental sciences 01 natural sciences computer vision Domain (software engineering) 3D detection 0202 electrical engineering electronic engineering information engineering 0105 earth and related environmental sciences Contextual image classification business.industry [INFO.INFO-CV]Computer Science [cs]/Computer Vision and Pattern Recognition [cs.CV] Pattern recognition computer vision machine learning domain adaptation LiDar 3D detection autonomous Driving Object (computer science) autonomous Driving Object detection Lidar machine learning Unsupervised learning RGB color model 020201 artificial intelligence & image processing Artificial intelligence business |
| Zdroj: | International Conference on 3D Vision International Conference on 3D Vision, Nov 2020, Fukuoka, Japan 3DV |
| Popis: | 3D object detectors based only on LiDAR point clouds hold the state-of-the-art on modern street-view benchmarks. However, LiDAR-based detectors poorly generalize across domains due to domain shift. In the case of LiDAR, in fact, domain shift is not only due to changes in the environment and in the object appearances, as for visual data from RGB cameras, but is also related to the geometry of the point clouds (e.g., point density variations). This paper proposes SF-UDA$^{3D}$, the first Source-Free Unsupervised Domain Adaptation (SF-UDA) framework to domain-adapt the state-of-the-art PointRCNN 3D detector to target domains for which we have no annotations (unsupervised), neither we hold images nor annotations of the source domain (source-free). SF-UDA$^{3D}$ is novel on both aspects. Our approach is based on pseudo-annotations, reversible scale-transformations and motion coherency. SF-UDA$^{3D}$ outperforms both previous domain adaptation techniques based on features alignment and state-of-the-art 3D object detection methods which additionally use few-shot target annotations or target annotation statistics. This is demonstrated by extensive experiments on two large-scale datasets, i.e., KITTI and nuScenes. Accepted paper at 3DV 2020 |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|