Workshop Program
09:00-09:20 |
Welcome & Opening Remarks |
09:20-10:00 |
Keynote: Thomas Funkhouser - Princeton University Learning Shape Templates using Structured Implicit Functions |
10:00-10:20 |
U-Net Based Convolutional Neural Network for Skeleton Extraction Oleg Panichev - Ciklum (Pixel SkelNetOn winner) |
10:20-10:45 |
Morning break |
10:45-11:25 |
Keynote: Ayellet Tal - Technion Shape Understanding for Archaeology |
11:25-11:45 |
|
11:45-13:00 |
Lunch break |
13:00-13:40 |
Keynote: Vincent Lepetit - University of Bordeaux 3D Registration of Rigid and Deformable Objects from Images |
13:40-14:00 |
|
14:00-14:40 |
Keynote: Leonidas Guibas - Stanford University Deep Learning for Point Cloud Data |
14:40-16:00 |
Poster session and afternoon break |
16:00-16:20 |
Parametric Skeleton Generation via Gaussian Mixture Models Wei Ke - Robotics Institute, Carnegie Mellon University (Parametric SkelNetOn winner) |
16:20-16:30 |
Award ceremony |
16:30-17:15 |
Collaboration panel (Leonidas Guibas, Thomas Funkhouser, Ayellet Tal, Vincent Lepetit) |
17:15-17:30 |
Closing & Walking to workshop dinner |
Computer vision approaches have made tremendous efforts toward understanding shape from various data formats, especially since entering the deep learning era. Although detection, recognition, and segmentation approaches achieve highly accurate results, there is less attention and research on extracting topological and geometric information from shapes. However, geometric representations provide a compact and intuitive abstractions for modeling, synthesis, compression, matching, and analysis. Extracting such representations are significantly different from segmentation and recognition tasks, as they condense both local and global information about the shape.
This workshop aims to bring together researchers from computer vision, computer graphics, and mathematics to advance the state of the art in topological and geometric shape analysis using deep learning. In addition to a traditional call for papers, our workshop proposes a challenge structured around shape understanding approaches in three domains. The datasets created and released for this competition will serve as reference benchmarks for future research in deep learning for shape understanding. Furthermore, different input and output data representations can become valuable testbeds for the design of robust computer vision and computational geometry algorithms, as well as understanding deep learning models built on representation in 3D and beyond.
We will also have an open submission format where the researchers can share their early work and novel (unpublished) research in geometric deep learning. Although we encourage all submissions to benchmark their results on the evaluation platform, there are other relevant research areas that our datasets do not address. For those areas, the scope of the submissions is enumerated but not limited to the following general topics:
Your submission must be written in English and must be sent in PDF format. Each submitted paper must be no longer than 4 pages excluding references. Please refer to the CVPR author submission guidelines for instructions regarding formatting, templates, and policies. The review process will be double blind but the papers will be linked to challenge submissions. Selected papers will be published in IEEE CVPRW proceedings, visible in IEEE Xplore and on CVF Website.
This workshop aims to bring together researchers from computer vision, computer graphics, and mathematics to advance the state of the art in topological and geometric shape analysis using deep learning. In addition to a traditional call for papers, our workshop proposes a challenge structured around shape understanding approaches in three domains. The datasets created and released for this competition will serve as reference benchmarks for future research in deep learning for shape understanding. Furthermore, different input and output data representations can become valuable testbeds for the design of robust computer vision and computational geometry algorithms, as well as understanding deep learning models built on representation in 3D and beyond.
We will also have an open submission format where the researchers can share their early work and novel (unpublished) research in geometric deep learning. Although we encourage all submissions to benchmark their results on the evaluation platform, there are other relevant research areas that our datasets do not address. For those areas, the scope of the submissions is enumerated but not limited to the following general topics:
- Boundary extraction from 2D/3D shapes
- Geometric deep learning on 3D and higher dimensions
- Generative methods for parametric representations
- Novel shape descriptors and embeddings for geometric deep learning
- Deep learning on non-Euclidean geometries
- Transformation invariant shape abstractions
- Shape abstraction in different domains
- Synthetic data generation for data augmentation in geometric deep learning
- Comparison of shape representations for efficient deep learning
- Applications of geometric deep learning in different domains
Your submission must be written in English and must be sent in PDF format. Each submitted paper must be no longer than 4 pages excluding references. Please refer to the CVPR author submission guidelines for instructions regarding formatting, templates, and policies. The review process will be double blind but the papers will be linked to challenge submissions. Selected papers will be published in IEEE CVPRW proceedings, visible in IEEE Xplore and on CVF Website.