FLAVR

Why we included this project

FLAVR is a flow-free frame interpolation model drawn from a WACV 2023 paper, and it stands out because a single forward pass can produce several intermediate frames at once. Rather than estimating optical flow, it learns to fill missing frames directly, which keeps it fast compared with older methods like SuperSloMo and DAIN while still delivering smooth high-frame-rate video. The repo ships pretrained 2x, 4x, and 8x models along with testing scripts, so you can apply slow-motion effects to your own clips without retraining. It also works as a readable codebase for studying spatio-temporal convolutions and channel gating, with training commands for Vimeo-90K and GoPro ready to extend.