GenEval 2: Addressing Benchmark Drift in Text-to-Image Evaluation

The article discusses GenEval 2, focusing on the issue of benchmark drift in text-to-image evaluation. This suggests a focus on improving the reliability and consistency of evaluating text-to-image models over time, as benchmarks can change and become less representative of actual model performance. The source being ArXiv indicates this is likely a research paper.