Reproduce “Latent Diffusion-GAN: Adversarial Learning in the Autoencoded Latent Space” 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
U-Chae Jun, Jaeeun Ko, Jiwoo Kang
LD-GAN reuses the pretrained VAE inside a latent diffusion model as an adversarial judge, scoring noisy and partly denoised latent codes through decoder–encoder reconstruction and multi-scale feature alignment. It fine-tunes the decoder without a separate discriminator and leaves diffusion sampling unchanged. In the reported Stable Diffusion 2.1 comparison, FID falls from 13.79 to 9.42 and CLIP score rises from 0.3098 to 0.3428 on sampled COCO2014 and Flickr30k pairs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.