Summary of Unified Triplet-level Hallucination Evaluation For Large Vision-language Models, by Junjie Wu et al.
Unified Triplet-Level Hallucination Evaluation for Large Vision-Language Models
by Junjie Wu, Tsz Ting Chung, Kai Chen, Dit-Yan Yeung
First submitted to arxiv on: 30 Oct 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); 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 proposed framework evaluates object and relation hallucination in Large Vision-Language Models (LVLMs) simultaneously, focusing on the relations between two objects. The framework assesses hallucinations on triplets extracted from LVLM responses, making it applicable to various vision-language tasks. A novel benchmark, Tri-HE, is introduced, demonstrating that relation hallucination is a more significant issue than object hallucination among existing LVLMs. To mitigate this problem, the authors propose a training-free approach, achieving comparable performance with GPT-4V and outperforming open-sourced counterparts on the Tri-HE benchmark. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large vision-language models can create fake information that doesn’t exist in an image. Most tests only check for object-related hallucinations, but this paper looks at relation hallucinations too – when models make up relationships between objects. The authors created a special test to evaluate both types of hallucinations together. They found that relation hallucinations are actually more common than object hallucinations and developed a way to fix the problem without needing to train the models further. |
Keywords
» Artificial intelligence » Gpt » Hallucination
