3DSAINT Representation for 3D Point Clouds

Chandra Kambhamettu; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 2765-2774

Abstract


This paper introduces a Sphere-based representation to model a 3D scene and show its performance on various tasks, including Structure from Motion (SfM) and 3D scene classification. A significant target application of this work is Mixed Reality, where 3D data can be efficiently represented, and synthetic and real data can be mixed for an immersive experience. Over the past few decades, 3D big data has garnered increased attention in computer vision. Acquiring, representing, reconstructing, querying, classifying, and visualizing 3D models for Mixed Reality has become crucial for many applications, such as medicine, architecture, entertainment, and bioinformatics. With the ever-increasing amount of data that the 3D scanners produce, storing, processing, and transmitting the data becomes challenging. Techniques that exploit the shape information need to be developed to model, classify and visualize the data. Our work offers a novel multi-scale surface representation based on spheres, with the ultimate goal of helping scientists to see and work with 3D data in Mixed Reality more effectively and efficiently. Om Sai Ram

Related Material


[pdf]
[bibtex]
@InProceedings{Kambhamettu_2023_CVPR, author = {Kambhamettu, Chandra}, title = {3DSAINT Representation for 3D Point Clouds}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2023}, pages = {2765-2774} }