KAIST Releases EDGAR: Diffusion Framework Refines Degraded Gaussian Head Avatars
KAIST · hf · 2026-11-06
KAIST's Visual Media Lab introduces EDGAR, a conditional diffusion framework with a fusion block and Mixture-of-Experts that refines degraded Gaussian head avatars.
- It injects self-supervised identity features and uses refined multi-view renderings as pseudo-targets to optimize a single avatar
- Training leverages synthetically generated failure datasets, and the paper shows recovery of avatars degraded at various levels from diverse sources
More from Multimodal
- AI-generated VHS-style zombie short 'Roxy & Brock' wows Reddit — brythehuman · 2026-11-06
- OpenMontage: open-source ComfyUI pipeline for consistent-character AI short dramas — Direct_Grab7063 · 2026-11-06
- Seedance 2.5 demo: creator says visuals and detail feel straight out of a blockbuster — SimplyAnnisa · 2026-11-06
- Seedance tier comparison: Mini to 2.5 spans $0.30-$2.00 per clip — Acrobatic-Result9667 · 2026-11-06
- Pika H3 + VOSR2.0 + RTX Super Resolution: 27-minute render on 8GB VRAM — solomon9892 · 2026-11-06
- Quick Test Pits gpt-image-2.6 Against Seedance 2.5 — DavidmComfort · 2026-11-06