25 sequences of real-world and synthetic data featuring high-temporal resolution event streams and synchronized RGB video frames. The collection provides the benchmark data for the CVPR 2021 TimeLens paper, focusing on reconstructing high-speed motion through event-based interpolation.
Use Cases
- Train neural networks to predict intermediate RGB frames using the `event_stream` and `boundary_frames` data.
- Benchmark event-based optical flow estimation by comparing predicted motion against the `ground_truth_frames`.
- Develop deblurring algorithms that utilize the high-temporal resolution of the `event_stream` to recover sharp images.
Strengths
- Contains event streams with (x, y, t, p) coordinates representing per-pixel intensity changes.
- Includes high-frame-rate ground truth RGB images for evaluating interpolation accuracy.
- Features 25 sequences from the HQF and HS-ERGB datasets captured with DAVIS240C and other event sensors.