Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward Modeling
For researchers evaluating image editing models and reward models, this provides a more challenging and realistic benchmark suite to better assess model capabilities and guide RL-based optimization.
Existing image editing benchmarks fail to reflect human judgment for frontier models due to limited difficulty and coarse evaluation, while reward model benchmarks use unrealistic settings. The authors introduce Edit-Compass (2,388 instances across six challenging tasks with fine-grained evaluation) and EditReward-Compass (2,251 preference pairs for realistic RL scenarios), demonstrating improved assessment capabilities.
Recent image editing models have achieved remarkable progress in instruction following, multimodal understanding, and complex visual editing. However, existing benchmarks often fail to faithfully reflect human judgment, especially for strong frontier models, due to limited task difficulty and coarse-grained evaluation protocols. In parallel, reward models have become increasingly important for RL-based image editing optimization, yet existing reward model benchmarks still rely on unrealistic evaluation settings that deviate from practical RL scenarios. These limitations hinder reliable assessment of both image editing models and reward models. To address these challenges, we introduce Edit-Compass and EditReward-Compass, a unified evaluation suite for image editing and reward modeling. Edit-Compass contains 2,388 carefully annotated instances spanning six progressively challenging task categories, covering capabilities such as world knowledge reasoning, visual reasoning, and multi-image editing. Beyond broad task coverage, Edit-Compass adopts a fine-grained multidimensional evaluation framework based on structured reasoning and carefully designed scoring rubrics. In parallel, EditReward-Compass contains 2,251 preference pairs that simulate realistic reward modeling scenarios during RL optimization.