Video Frame Interpolation via Cyclic Fine-Tuning and Asymmetric Reverse Flow

Morten Hannemose*, Janus Nørtoft Jensen, Gudmundur Einarsson, Jakob Wilm, Anders Bjorholm Dahl, Jeppe Revall Frisvad

*Kontaktforfatter for dette arbejde

Publikation: Kapitel i bog/rapport/konference-proceedingKonferencebidrag i proceedingsForskningpeer review

22 Downloads (Pure)


The objective in video frame interpolation is to predict additional in-between frames in a video while retaining natural motion and good visual quality. In this work, we use a convolutional neural network (CNN) that takes two frames as input and predicts two optical flows with pixelwise weights. The flows are from an unknown in-between frame to the input frames. The input frames are warped with the predicted flows, multiplied by the predicted weights, and added to form the in-between frame. We also propose a new strategy to improve the performance of video frame interpolation models: we reconstruct the original frames using the learned model by reusing the predicted frames as input for the model. This is used during inference to fine-tune the model so that it predicts the best possible frames. Our model outperforms the publicly available state-of-the-art methods on multiple datasets.

TitelImage Analysis - Proceedings of the 21st Scandinavian Conference, SCIA 2019
RedaktørerMichael Felsberg, Per-Erik Forssén, Jonas Unger, Ida-Maria Sintorn
ForlagSpringer VS
ISBN (Trykt)9783030202040
StatusUdgivet - 2019
Begivenhed21st Scandinavian Conference on Image Analysis, SCIA 2019 - Norrköping, Sverige
Varighed: 11. jun. 201913. jun. 2019


Konference21st Scandinavian Conference on Image Analysis, SCIA 2019
NavnLecture Notes in Computer Science


Dyk ned i forskningsemnerne om 'Video Frame Interpolation via Cyclic Fine-Tuning and Asymmetric Reverse Flow'. Sammen danner de et unikt fingeraftryk.