Summary of Doctr: Disentangled Object-centric Transformer For Point Scene Understanding, by Xiaoxuan Yu et al.
DOCTR: Disentangled Object-Centric Transformer for Point Scene Understanding
by Xiaoxuan Yu, Hao Wang, Weiming Li, Qiang Wang, Soonyong Cho, Younghun Sung
First submitted to arxiv on: 25 Mar 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 The proposed Disentangled Object-Centric TRansformer (DOCTR) tackles the challenging task of point scene understanding by simultaneously segmenting, estimating pose, and reconstructing mesh for each object in a unified manner. Unlike existing methods that process objects independently with multiple stages, DOCTR leverages object-centric representations to facilitate learning and optimize relationships between multiple objects. The novel approach introduces semantic-geometry disentangled query (SGDQ) design, which separates features attending to semantic and geometric information relevant to sub-tasks. A hybrid bipartite matching module is employed for training. Experimental results demonstrate DOCTR’s state-of-the-art performance on the challenging ScanNet dataset. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The researchers developed a new way to understand 3D scenes by dividing them into individual objects, estimating each object’s position, and creating detailed models of each one. This approach helps computers better recognize relationships between objects in a scene. The team created a special type of representation that allows computers to focus on the right details for each task, like recognizing shapes or understanding what’s happening in a scene. They tested their method on a difficult dataset and found it outperformed other methods. |
Keywords
» Artificial intelligence » Scene understanding » Transformer
