Summary of Polos: Multimodal Metric Learning From Human Feedback For Image Captioning, by Yuiga Wada et al.
Polos: Multimodal Metric Learning from Human Feedback for Image Captioning
by Yuiga Wada, Kanta Kaneda, Daichi Saito, Komei Sugiura
First submitted to arxiv on: 28 Feb 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 The proposed paper introduces a novel automatic evaluation metric for image captioning models called Polos. This metric is designed to closely align with human judgments and outperform existing metrics in handling hallucinations and generalizing across diverse images and texts. To develop Polos, the authors introduce Multimodal Metric Learning from Human Feedback (M^2LHF), a framework that leverages large-scale contrastive learning to train embeddings. The authors also construct the Polaris dataset, comprising 131K human judgments from 550 evaluators, which is ten times larger than standard datasets. The proposed approach achieves state-of-the-art performance on several benchmark datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Image captioning models need a reliable way to evaluate how well they understand images and create accurate captions. A new metric called Polos helps measure this by comparing the generated captions with human-written ones. To make sure Polos is fair and works well, the authors train it using large amounts of data and ask many people for their opinions. The result is a more accurate way to evaluate image captioning models. |
Keywords
» Artificial intelligence » Image captioning
