Summary of Can Shape-infused Joint Embeddings Improve Image-conditioned 3d Diffusion?, by Cristian Sbrolli et al.
Can Shape-Infused Joint Embeddings Improve Image-Conditioned 3D Diffusion?
by Cristian Sbrolli, Paolo Cudrano, Matteo Matteucci
First submitted to arxiv on: 2 Feb 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 recent fusion of CLIP and DDPMs has led to groundbreaking text-to-image generation capabilities. However, this study questions whether CLIP is the optimal choice for image-to-shape generation. To address this, CISP (Contrastive Image Shape Pre-training) is introduced as a framework that aligns 2D images with 3D shapes in a shared embedding space, capturing 3D characteristics overlooked by text-image focused CLIP models. This study compares CISP’s guidance performance against CLIP guided models, evaluating generation quality, diversity, and coherence of produced shapes with conditioning images. The results show that while matching CLIP in terms of generation quality and diversity, CISP significantly improves coherence with input images, highlighting the value of incorporating 3D knowledge into generative models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary CISP is a new way to make computers create 3D shapes from pictures. Currently, there are special models that can take words and turn them into pictures. But what if we want to turn pictures into 3D shapes? This study asks if the same model is good for both tasks. They created a new model called CISP that helps computers understand how to create 3D shapes from pictures. They tested this model against other models and found that it does a better job of creating shapes that look like what they should look like based on the picture. |
Keywords
» Artificial intelligence » Embedding space » Image generation
