AI
DC-SAE: Deep Compression Semantic Autoencoder for Faster Diffusion Convergence
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 29/09/2026, 20:00:00 UTC
Briefing created: 01/10/2026, 09:06:22 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 03:18:25 UTC.
Briefings
This article presents DC-SAE, a Decoupled Compact Semantic Autoencoder designed to balance high-compression tokenization with efficient diffusion training. By combining macro-level semantic encoders with pixel-level detail preservation, DC-SAE achieves 32x spatial compression on ImageNet while maintaining high PSNR and gFID scores. The model enables a 1.6B-parameter DiT to generate high-quality images at 1024x1024 resolution with significantly faster diffusion convergence compared to previous baselines.
Key points
- DC-SAE achieves 32x spatial compression on ImageNet with 29.79 PSNR and 3.37 gFID, outperforming DC-AE by 13.5% and 54.9% on PSNR and gFID respectively.
- A 1.6B-parameter DiT using DC-SAE achieves 0.84 on GenEval and 86.007 on DPG-Bench at 1024x1024 resolution, enabling faster diffusion model convergence.
Context and limitations
The study is limited to ImageNet and DiT models; results on other datasets or architectures remain unverified.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 01/10/2026, 09:06:22 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeibqnytz44zwnafhrfe5grda5kc6upob6diyky25kxnft6w2xjak7m