[KongchangAI]
Benchmark

ImgEdit

A benchmark for evaluating instruction-driven image editing capabilities, covering multiple editing types such as object replacement, addition, removal and isolation. It scores performance by combining instruction-following fidelity and image preservation quality.

Timeline (last 90 days)

Oct 12

On Qwen-Image-Edit, Rubric-CEPR improved ImgEdit from 4.36 to 4.60, a gain of +5.5%.

Unverified50%
Oct 12

On Qwen-Image-Edit, Rubric-CEPR achieved a +24.9% improvement on the object isolation task.

Unverified50%
Oct 12

When the same pipeline was applied to Step1X-Edit, a +7.8% improvement was achieved on ImgEdit.

Unverified50%
Oct 12

ImgEdit, GEdit-Bench and Complex-Edit are all benchmark test sets for evaluating instruction-driven image editing capability.

Unverified50%

All Facts (4)

Source Articles