April 30, deep learning for shape synthesis

Slides:

24 Data-driven shape modeling - deep learning, shape synthesis.pdf Download 24 Data-driven shape modeling - deep learning, shape synthesis.pdf

Panopto recording:

Learning check:

For the learning check, read the paper "DISN: Deep Implicit Surface Network for Links to an external site.
High-quality Single-view 3D Reconstruction Links to an external site." by Wang et al. (NeurIPS 2019, code Links to an external site.) and answer the questions below:

• Describe the problem statement that the problem addresses.
• Describe the shape representation that the network uses. List two advantages of their approach over previous techniques.
• Describe the  motivation for using local features (see Fig. 4) in their pipeline. How is the camera pose estimation used to obtain local features?
• Given the trained network and an input image, describe the steps that are performed to reconstruct a 3D shape.
• Formulate one question that you have about the paper.

Q&A

Here is the feedback to your questions from the learning check:
https://docs.google.com/document/d/1m610CHHPvNieEDfkJyYfqw1uBQMMP_ZgADz7M8fk1JM Links to an external site.