Frame Interpolation
Generating intermediate frames between two video frames to create smooth slow-motion or increase frame rate.
Think of it like
Flip-book animation: you draw frames 1 and 3, and fill in frame 2 to make it smooth.
Example
A video at 30fps -> interpolate to 60fps or 120fps for slo-mo sports. A 2-frame clip -> generate 10 frames in between.
How it actually works
Requires predicting motion (optical flow) and synthesizing pixels. Challenges: occlusion (new areas becoming visible), disocclusion (disappearing areas), blur, and ambiguity (multiple plausible in-betweens). Modern methods: flow-based (predict motion, warp + inpaint) or learning-based end-to-end (RIFE, trained on large video datasets). Diffusion-based approaches emerging. Must preserve temporal coherence.
For product teams
Powers video upsampling, slow-motion effects, frame rate conversion for streaming.
For engineers
Datasets: Vimeo90K, YouTube, custom. Metric: PSNR, SSIM, perceptual loss. Models: RIFE, DAIN.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome