US EditionPublished • loading... • Updated
Hacking Reference-Free Image Captioning Metrics: Uncovering Hidden Flaws
Summary by Newswise
This story is only covered by news sources that have yet to be evaluated by the independent media monitoring agencies we use to assess the quality and reliability of news outlets on our platform. Learn more here.1 Articles
1 Articles
Hacking Reference-Free Image Captioning Metrics: Uncovering Hidden Flaws
Addressing hidden flaws in VLM-based reference-free image captioning metrics, researchers from Nanjing University employed a reinforcement learning "hacking" strategy to test mainstream evaluation tools. They discovered that while these metrics correlate with human judgment, they can be manipulated to favor unreadable, incoherent text. To solve this, the team introduced Negative Text Contrastive Learning (NTCL), which significantly improves metr…
·Charlottesville, United States
Read Full ArticleCoverage Details
Total News Sources1
Leaning Left0Leaning Right0Center0Last UpdatedBias DistributionNo sources with tracked biases.
Bias Distribution
- There is no tracked Bias information for the sources covering this story.
Factuality
To view factuality data please Upgrade to Premium
