EDGAR Refines Gaussian Avatars via Diffusion
Original title: EDGAR: Efficient Diffusion-Driven Gaussian Avatar Refinement via Synthetic Failure
EDGAR corrects artifacts in Gaussian avatar renderings using a fusion block with Sapiens identity features and a Mixture-of-Experts architecture.
Recent Gaussian splatting methods suffer from under-constrained geometry and artifacts in monocular avatar reconstruction. EDGAR is a conditional diffusion framework that refines degraded frames by injecting identity features into a one-step diffusion backbone. It constructs EDGAR-SET with hybrid triplets simulating failures and optimizes a single Gaussian avatar using refined multi-view renderings as pseudo-targets.
- EDGAR-SET pairs degraded renderings with ground-truth images
- Fusion block enables artifact removal without efficiency loss
- MoE addresses diverse degradation types
- Improvements in perceptual and reconstruction metrics
- Enhanced calibrated consistency across views
From the original article
1 min read at huggingface.co
Recent advances in Gaussian splatting have enabled efficient and photorealistic head avatar reconstruction, yet monocular video-based or image-based methods still suffer from under-constrained geometry and texture, leading to artifacts such as blurring, stretching, and temporal instability.
By Kwan Yun, Minhyeok Bang, Colette Suhjung Koo · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.