Researchers have developed a new image restoration model called Flexible Image Transformer (FIT). FIT explicitly models degradation awareness throughout its entire pipeline, from patch sampling to pixel reconstruction, unlike previous methods that only injected task/degradation conditions after tokenization. The model uses a degradation encoder to predict a global degradation vector and a spatial degradation map, which adaptively condition patch embedding and unembedding. FIT achieves state-of-the-art performance on five standard benchmarks, outperforming recent unified restoration methods by up to 1.1 dB. AI
IMPACT This new model advances image restoration capabilities by offering a more robust and adaptable approach to handling diverse degradation types.
RANK_REASON Publication of a new research paper detailing a novel model and its benchmark performance. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
