Reproduce “Deep Back-Projection Networks For Super-Resolution” 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
Muhammad Haris, Greg Shakhnarovich, Norimichi Ukita
The Deep Back-Projection Network (DBPN) introduces an architecture for single image super-resolution that employs iterative up and down-sampling stages to provide mutual error feedback. This method achieves state-of-the-art performance on various benchmarks, particularly for large upscaling factors, while maintaining model efficiency.
Key insights
- Iterative up and down-sampling stages, inspired by classical back-projection, allow the network to learn complex relationships and refine features by projecting prediction errors.
- Mutual feedback between low-resolution and high-resolution feature spaces through dedicated projection units is crucial for error correction and better reconstruction.
- Dense connections within the projection units facilitate feature reuse and gradient flow, enhancing the network's representational power and allowing for deeper error correction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.