Summary of Finetuning Pre-trained Model with Limited Data For Lidar-based 3d Object Detection by Bridging Domain Gaps, By Jiyun Jang et al.
Finetuning Pre-trained Model with Limited Data for LiDAR-based 3D Object Detection by Bridging Domain Gaps
by Jiyun Jang, Mincheol Chang, Jongwon Park, Jinkyu Kim
First submitted to arxiv on: 2 Oct 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Robotics (cs.RO)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes Domain Adaptive Distill-Tuning (DADT), a novel method for adapting LiDAR-based 3D object detectors to new domains with limited target data. The approach builds upon recent studies that demonstrate the effectiveness of pre-training backbones using self-supervised learning on large-scale unlabeled LiDAR frames. However, these models often struggle to generalize well without substantial amounts of target domain data. DADT addresses this challenge by introducing regularizers that align object-level and context-level representations between a pre-trained model and its finetuned counterpart in a teacher-student architecture. The authors evaluate their method on driving benchmarks such as the Waymo Open dataset and KITTI, achieving significant gains in accuracy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us make LiDAR-based 3D object detectors better at recognizing objects in new environments with different sensors or locations. Usually, we need to collect and label data for these new setups, which can be time-consuming and expensive. Researchers have shown that pre-training models on large amounts of unlabeled LiDAR frames is helpful. However, even with this training, the models often struggle to adapt well without more target domain data. To fix this, the authors propose a new method called Domain Adaptive Distill-Tuning (DADT). They use special regularizers to help the pre-trained model and its finetuned version agree on what’s important in the data. The results show that DADT helps improve accuracy when detecting objects in driving scenarios. |
Keywords
» Artificial intelligence » Self supervised
