BiT: Real-World Motion Blur Interpolation Sequences for CVPR 2023

Real-world motion blur sequences captured via a beam-splitter system support the Blur Interpolation Transformer (BiT) presented at CVPR 2023. Created by zzh-tech, the data enables the conversion of single blurred images into sharp video sequences.

Use Cases

• Training video-frame-interpolation models to reconstruct sharp sequences from blurred image inputs
• Benchmarking image-to-video restoration performance on real-world motion trajectories
• Evaluating low-level vision transformers on beam-splitter captured video enhancement tasks

Strengths

• Peer-reviewed origin from CVPR 2023
• Real-world data capture using a beam-splitter system
• MIT license for open research use

Limitations

• Unknown total record count and storage size
• Potential hardware-specific bias from the beam-splitter capture method

Provenance

Source

CVPR 2023 paper 'Blur Interpolation Transformer for Real-World Motion from Blur'

Collection Method

Captured using a beam-splitter system to record simultaneous blurred and sharp frames

Freshness

Last updated March 2024.