Omni IIE Bench
Benchmarking the Practical Capabilities of Instruction-based Image Editing Models
[arXiv] [HuggingFace Dataset] [Project Page]
Overview
Omni IIE Bench is a human-annotated benchmark for diagnosing editing consistency of Instruction-based Image Editing (IIE) models. Unlike existing benchmarks that use mixed evaluations, Omni IIE Bench employs a dual-track diagnostic design with a 16-turn dialogue length — the longest among all IIE benchmarks.… See the full description on the dataset page: https://huggingface.co/datasets/YamJoy/OmniIIEBench.
