BlurRF-Synth
3DImagesMITIntroduced 2025-02-20
The first large-scale dataset for training and evaluating novel-view synthesis from blurred images.
The dataset comprises a train set and a test set for each type of blur: camera motion blur and defocus blur. The train set includes 65 scenes, each containing 29 pairs of synthetically blurred images and their corresponding sharp images, captured from different viewpoints. The test set includes 10 scenes, each with 29 blurred-sharp image pairs and five additional sharp images from different viewpoints for evaluating novel-view synthesis quality.