Summary of Geogen: Geometry-aware Generative Modeling Via Signed Distance Functions, by Salvatore Esposito et al.
GeoGen: Geometry-Aware Generative Modeling via Signed Distance Functions
by Salvatore Esposito, Qingshan Xu, Kacper Kania, Charlie Hewitt, Octave Mariotti, Lohit Petikam, Julien Valentin, Arno Onken, Oisin Mac Aodha
First submitted to arxiv on: 6 Jun 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 We introduce a novel approach for generating 3D geometry and images from single-view collections, addressing limitations of existing methods. The proposed model, GeoGen, is an SDF-based 3D generative model trained end-to-end. Initially, we reinterpret volumetric density as an SDF, allowing us to incorporate useful priors for generating valid meshes. However, these priors hinder the model’s ability to learn details, limiting its applicability to real-world scenarios. To alleviate this issue, we make the transformation learnable and constrain the rendered depth map to be consistent with the zero-level set of the SDF through adversarial training. Our experiments on multiple datasets demonstrate that GeoGen produces better geometry than previous generative models based on neural radiance fields. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine being able to create 3D images from just one view, like a single photo. This is what our new approach, called GeoGen, can do. We’re trying to solve the problem of noisy and unconstrained generated geometry that previous methods have struggled with. To fix this, we created a new model that learns to generate valid meshes while still including details. We tested our approach on multiple datasets and found that it produces better results than other similar methods. |
Keywords
» Artificial intelligence » Generative model
