Reproduce “Self-Similarity Priors: Neural Collages as Differentiable Fractal Representations” 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

Michael Poli, Winnie Xu, Stefano Massaroli, Chenlin Meng, Kuno Kim, Stefano Ermon

Neural Collage compresses images by using a trained encoder to predict parameters for a self-referential patch transformation. The transformation blends transformed source patches into output patches and is iterated to reconstruct the image. On DOTA images, it reaches 32.12 dB at 0.31 bits per pixel, above COIN at a similar rate, and encodes 100 times faster than PIFS after training; block-DCT has higher PSNR.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.