Summary of Improving Composed Image Retrieval Via Contrastive Learning with Scaling Positives and Negatives, by Zhangchi Feng et al.
Improving Composed Image Retrieval via Contrastive Learning with Scaling Positives and Negatives
by Zhangchi Feng, Richong Zhang, Zhijie Nie
First submitted to arxiv on: 17 Apr 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 proposed Composed Image Retrieval (CIR) method addresses the limitations of existing approaches by generating more positive and negative examples using a multi-modal large language model. A two-stage fine-tuning framework is designed to optimize the representation space rapidly, introducing plenty of static representations of negatives in the second stage. This plug-and-play approach can be easily applied to existing CIR models without changing their original architectures. The method achieves state-of-the-art results on both FashionIQ and CIRR datasets and performs well in zero-shot composed image retrieval, providing a new solution for low-resources scenarios. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine you’re trying to find a specific picture by describing it to a computer. This paper develops a way to make this process more efficient by creating more examples of the correct answers (positive examples) and incorrect ones (negative examples). They use a special language model to generate these examples, which can be used with existing image retrieval models without changing them. The new approach works well on two different datasets and even performs well when there’s no training data available. |
Keywords
» Artificial intelligence » Fine tuning » Language model » Large language model » Multi modal » Zero shot
