Reproduce “Low-Resolution Editing is All You Need for High-Resolution 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
Junsung Lee, Hyunsoo Lee, Yong Jae Lee, Bohyung Han
Researchers from Seoul National University and the University of Wisconsin-Madison introduced ScaleEdit, a framework that enables high-resolution image editing by integrating fine-grained detail transfer from a source image directly into the editing process. This method achieved superior performance compared to super-resolution baselines, showing the highest HaarPSI (0.342) and lowest LPIPS (0.460) for 1K-editing tasks.
Key insights
- Directly integrating fine-grained detail transfer from the high-resolution source into the editing process is essential, as post-processing super-resolution cannot adequately recover lost details.
- Patch-wise optimization, when combined with a specific detail enhancement module and a robust synchronization strategy, enables effective high-resolution editing while preserving global consistency.
- A lightweight 1x1 convolutional transfer function, optimized to align generation trajectories, is sufficient for spatially adaptive detail injection without extensive model retraining.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.