Reproduce “Deep Image Harmonization” 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
Yi-Hsuan Tsai, Xiaohui Shen, Zhe Lin, Kalyan Sunkavalli, Xin Lu, Ming-Hsuan Yang
A collaboration between UC Merced and Adobe Research introduces the first end-to-end deep convolutional neural network for image harmonization, adjusting foreground appearances to blend seamlessly with new backgrounds. This deep learning approach achieves superior visual quality and significantly faster processing speeds, enabling real-time operation compared to traditional methods.
Key insights
- Generating large-scale, high-quality training data for image manipulation tasks can be achieved by algorithmically creating 'unharmonized' inputs from naturally consistent ground truth images.
- Incorporating semantic understanding through a jointly trained scene parsing branch can significantly improve the quality and realism of image harmonization.
- An end-to-end deep learning framework can effectively replace complex multi-stage traditional methods, leading to substantial gains in both output quality and computational efficiency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.