ICE-Bench
收藏资源简介:
ICE-Bench是由阿里巴巴集团推出的统一和全面的图像生成评估基准。该数据集涵盖了31个细粒度的生成任务,分为无参考/有参考的图像创建/编辑四大类别,旨在全面评估图像生成模型的能力。数据集由真实场景和虚拟生成的图像混合构成,有效提高了数据多样性,减轻了模型评估中的偏差问题。它包含了6538个任务实例,旨在解决图像生成模型自动评估的挑战,推动该领域的研究进展。
ICE-Bench is a unified and comprehensive image generation evaluation benchmark launched by Alibaba Group. This dataset encompasses 31 fine-grained generation tasks, categorized into four groups: reference-free and reference-based image creation and editing, with the goal of comprehensively evaluating the capabilities of image generation models. The dataset comprises a mix of real-world scenes and virtually generated images, effectively boosting data diversity and mitigating bias issues in model evaluation. It contains 6538 task instances, aiming to address the challenges of automatic evaluation of image generation models and advance research progress in this field.

- 1ICE-Bench: A Unified and Comprehensive Benchmark for Image Creating and Editing阿里巴巴集团 · 2025年



