Researchers have introduced two new datasets and models for real-world image deblurring. The first, "Deblurring in the Wild," utilizes smartphone slow-motion videos to create a large dataset of over 42,000 blur-sharp image pairs, significantly larger than existing datasets. This dataset highlights performance degradation in current state-of-the-art models, indicating a need for more robust solutions. The second contribution, RealVDeblur, presents an efficient generative framework using a one-step diffusion model and 3D Gaussian Splatting for realistic blur synthesis, aiming for generalizable video deblurring that improves downstream applications like 3D reconstruction. AI
IMPACT Advances in deblurring could improve image quality in consumer devices and enhance downstream computer vision tasks like 3D reconstruction.
RANK_REASON Two research papers introducing new datasets and models for image deblurring.
- 3D Gaussian Splatting
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Deblurring in the Wild
- Gotit.pub
- Hugging Face
- RealVDeblur
- ScienceCast
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
