A source of video sequences and synthetic data for the DeFMO framework, focusing on the deblurring and shape recovery of fast-moving objects (FMOs). It includes blurred input frames paired with high-speed ground truth sequences to facilitate the reconstruction of sharp object appearance and sub-frame trajectories.
Use Cases
- Train neural networks for temporal super-resolution using high-speed ground truth frame sequences
- Develop deblurring algorithms that recover object shape from motion-blurred input images
- Benchmark trajectory estimation models by comparing predicted sub-frame positions against ground truth motion paths
Strengths
- Includes high-speed ground truth sequences for evaluating temporal super-resolution performance
- Features blurred frames where objects exhibit significant motion blur relative to their size
- Provides object masks and appearance ground truth for shape recovery tasks