How Kuaishou’s KFRUC Algorithm Achieves Seamless Slow‑Motion Video

What is Video Frame Interpolation?

Video frame interpolation inserts intermediate frames between original frames by analyzing motion, increasing frame rate and smoothing playback. Higher FPS yields a smoother visual experience. The technique is also used in animation production and video coding standards such as VVC’s DMVR.

Types of Frame Interpolation Algorithms

There are three main categories:

Simple duplication or blending of frames, which does not improve motion continuity.

Motion estimation and motion compensation (MEMC), which generates motion‑consistent frames and is used in commercial products—Kuaishou’s KFRUC belongs to this class.

Deep‑learning‑based methods that predict intermediate frames but require high computation and may be unstable in complex scenes.

Kuaishou’s KFRUC Algorithm

KFRUC (Kwai Frame Rate Up‑Conversion) divides each frame into many blocks, then performs motion estimation, occlusion detection, and motion compensation to synthesize intermediate frames.

Motion Estimation

KFRUC uses a hybrid motion‑estimation approach with three candidate vectors per block and selects the best one. It includes:

Bilateral (bidirectional) search : matches symmetric blocks in the preceding and following reference frames. If the match is reliable, no further search is needed.

Forward search : used when the bidirectional search fails due to occlusion, finding a matching block in the later reference frame and its counterpart in the earlier frame.

Backward search : analogous to forward search but starts from the later reference frame.

Occlusion Detection

Blocks are classified as front‑occluded, rear‑occluded, or normal based on which reference frame provides a better match. This classification guides the subsequent compensation step.

Motion Compensation

According to the occlusion class, KFRUC selects reference blocks from the appropriate frame: rear‑occluded blocks use the earlier frame, front‑occluded blocks use the later frame, and normal blocks use the average of both. The algorithm also applies OBMC deblocking to improve visual quality.

For challenging scenes with fast motion or heavy blur, KFRUC may fall back to averaging the two surrounding frames instead of inserting a synthesized one, avoiding visual artifacts.

Results and Outlook

After motion estimation, occlusion handling, and compensation, KFRUC produces motion‑continuous intermediate frames, enabling smooth slow‑motion playback without stutter.

The team notes ongoing challenges such as complex non‑linear motion and lighting changes, and they plan to explore neural‑network engines and the latest deep‑learning advances to develop higher‑quality interpolation algorithms for an even better visual experience.