Researchers have created a large-scale dataset called DF3DV-1K to help improve the development of 'distractor-free' vision systems. This dataset includes over 89,000 images from 1,048 scenes, with both clean and cluttered versions of each scene. The team used this dataset to benchmark nine recent methods for generating radiance fields, identifying which ones are most robust in challenging scenarios. They also demonstrated an application of the dataset by fine-tuning a diffusion-based image enhancer to improve these methods. --- Why it matters: This matters because it provides a large-scale dataset and benchmark for evaluating distractor-free vision systems, which can help advance the field beyond scene-specific approaches. It's particularly relevant to researchers working on novel view synthesis and radiance fields. Source: https://arxiv.org/abs/2604.13416