Reproduce “CheckDistill: Adaptive Checklist Distillation for Efficient Distributional Reward Modeling in Image Editing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yaqi Cai, Zuan Gao, Mengting Chen, Wei Li, Bo Zheng
CheckDistill trains a direct reward model for instruction-guided image editing using instance-specific checklists as privileged evidence. The student learns to predict four scores without generating checklists at inference. On MMRB2, CheckDistill with Qwen3.5-9B reaches 72.30% overall accuracy versus SpatialReward’s 66.10%, using about 10 output tokens; this measures output tokens, not elapsed time or total computation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.