Summary of Updating Clip to Prefer Descriptions Over Captions, by Amir Zur et al.
Updating CLIP to Prefer Descriptions Over Captions
by Amir Zur, Elisa Kreiss, Karel D’Oosterlinck, Christopher Potts, Atticus Geiger
First submitted to arxiv on: 12 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
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 This paper aims to improve CLIPScore, a generic metric for evaluating text-image similarity, by enhancing its ability to distinguish between complementary captions and descriptive texts. The current CLIPScore model fails to capture this distinction, which is crucial in scenarios like accessibility where images are replaced with descriptions. To address this limitation, the authors update the CLIP model using the Concadia dataset, a novel approach that leverages parameter-efficient fine-tuning and a loss objective inspired by causal interpretability research. The resulting model not only correlates well with human judgements from blind and low-vision individuals but also maintains its transfer capabilities and exhibits interpretable structure. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine you’re trying to describe an image to someone who can’t see it. A new way of measuring how well a text matches an image is needed, as the current method can’t tell if the text is meant to help or replace the image. The researchers created a better model by fine-tuning an existing one using special data and a specific goal. This improved model is good at matching texts with images and also works well when transferring to new situations. It even helps people who are blind or have low vision understand how well a text matches an image. |
Keywords
» Artificial intelligence » Fine tuning » Parameter efficient
