DF3DV-1K: A Large-Scale Dataset and Benchmark for Distractor-Free Novel View Synthesis
Researchers have introduced DF3DV-1K, a large-scale dataset designed to advance research in distractor-free novel view synthesis. This dataset comprises over 1,000 scenes, each with both clean and cluttered image sets, totaling nearly 90,000 images captured with consumer cameras. It includes 128 types of distractors and 161 scene themes, with a specific subset, DF3DV-41, curated for evaluating method robustness. The dataset has been used to benchmark existing radiance field methods and 3D Gaussian Splatting, and to fine-tune a diffusion-based enhancer, showing significant improvements in rendering quality. AI
IMPACT Facilitates development of more robust and generalizable novel view synthesis techniques beyond scene-specific approaches.