Summary of Sc3d: Label-efficient Outdoor 3d Object Detection Via Single Click Annotation, by Qiming Xia et al.
SC3D: Label-Efficient Outdoor 3D Object Detection via Single Click Annotation
by Qiming Xia, Hongwei Lin, Wei Ye, Hai Wu, Yadan Luo, Cheng Wang, Chenglu Wen
First submitted to arxiv on: 15 Aug 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 SC3D is an innovative method for training LiDAR-based outdoor 3D object detectors using only a single coarse click on the bird’s eye view of the 3D point cloud. This label-efficient approach eliminates the need for expensive bounding box annotations, reducing annotation costs by up to 99.8%. SC3D adopts a progressive pipeline that includes mixed pseudo-label generation and mix-supervised teacher and student networks. These networks leverage limited click annotations to generate mixed supervision information, allowing the detector to learn unclicked instances on popular datasets like nuScenes and KITTI. Experimental results demonstrate that SC3D achieves state-of-the-art performance in weakly-supervised 3D detection tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary SC3D is a new way to teach machines to detect objects from LiDAR data without needing a lot of information about the objects’ shapes. This makes it much cheaper and more efficient than current methods. The team created a system that uses a single rough click on the object’s location in 2D space to train the machine learning model. They also developed special algorithms that help the model learn from these limited annotations. When tested on real-world data, SC3D performed better than other methods that use similar approaches. |
Keywords
» Artificial intelligence » Bounding box » Machine learning » Supervised
