15,000 image pairs consisting of distorted rolling shutter (RS) frames and their corresponding sharp global shutter (GS) ground truth. The collection covers diverse dynamic scenes with varying degrees of motion blur and geometric distortion generated from high-frame-rate video sequences.
Use Cases
- Train deep learning architectures to map distorted RS_image inputs to sharp GS_ground_truth outputs
- Benchmark joint rolling shutter correction and deblurring (RSCD) algorithms against provided global shutter references
- Analyze the impact of sensor readout timing on image restoration quality using the temporal sequences
Strengths
- 15,000 high-resolution image pairs of rolling shutter and global shutter frames
- Synthetic data generated from 1000fps high-speed video to ensure pixel-accurate ground truth
- Includes complex motion patterns including camera ego-motion and independent object movement