Summary of Domain Adaptation with a Single Vision-language Embedding, by Mohammad Fahes et al.
Domain Adaptation with a Single Vision-Language Embedding
by Mohammad Fahes, Tuan-Hung Vu, Andrei Bursuc, Patrick Pérez, Raoul de Charette
First submitted to arxiv on: 28 Oct 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
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 new framework for domain adaptation uses a single Vision-Language latent embedding to adapt source data to target conditions without requiring access to full target data. This is achieved by leveraging a contrastive language-image pre-training model, which proposes prompt/photo-driven instance normalization (PIN). PIN mines multiple visual styles using the VL latent embedding, optimizing affine transformations of low-level source features. The framework can utilize language prompts describing the target domain, partially optimized language prompts, or single unlabeled target images to generate the VL embedding. This approach enables zero-shot and one-shot unsupervised domain adaptation for tasks like semantic segmentation, outperforming relevant baselines in both settings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper presents a new way to adapt source data to target conditions without needing all the target data. It uses a special kind of map that connects language and images, called Vision-Language. This map helps generate many different visual styles from just one piece of information about the target domain. The approach is useful for adapting source data to target domains without any labeled data, which can be hard to come by in some situations. |
Keywords
» Artificial intelligence » Domain adaptation » Embedding » One shot » Prompt » Semantic segmentation » Unsupervised » Zero shot
