An Efficient Quality Metric for Video Frame Interpolation
Video Frame Interpolation (VFI) enhances temporal video quality for applications like slow-motion effects and broadcast frame-rate conversion. While modern VFI methods that use optical flow and deep networks can handle complex motion and occlusions, evaluating the quality of interpolated content remains challenging. Traditional metrics like peak signal-to-noise ratio (PSNR) and Stuctural Similarity Index (SSIM) ignore temporal information, while perceptual metrics like Learned Perceptual Image Patch Similarity (LPIPS) focus only on spatial aspects. Recent VFI-specific metrics, such as FloLPIPS, incorporate motion-field errors to detect temporal inconsistencies but are computationally expensive, limiting their practical use in training or real-time assessment. Building on previous work in motion picture restoration, we developed PSNRDIV, which uses divergence to detect irregularities in motion fields. As such, we can identify problematic motion field patterns that degrade the quality of interpolated frames. PSNRDIV requires only one sequence's motion field, thereby reducing computational load. Evaluation on the Bristol Video Intelligence (BVI)-VFI dataset shows statistically significant improvements over FloLPIPS. It achieves these gains while being 2.5x faster and using 6x less memory.
- Print ISSN
- 1545-0279
- Electronic ISSN
- 2160-2492
- Published
- 2026-07
- Content type
- Original Research
- Keywords
- video frame interpolation quality, temporal consistency metrics
- DOI
- 10.5594/JMI.2026/WZPJ1373