search

hzwer / ECCV2022-RIFE

ECCV2022 - Real-Time Intermediate Flow Estimation for Video Frame Interpolation
MIT License
4.41k stars 438 forks source link

Question about using Pytorch-LiteFlownet during retraining/fine-tuning #248

Closed jayrmh closed 2 years ago

jayrmh commented 2 years ago

Hey guys! I love your work, and I am working on a survey paper exploring interpolation algorithms in medical domain. So far your method is one of the best performing!

I would to fine-tune the RIFE HDv3 model on my dataset which has image triplets. I have also gotten Ptroch-LiteFlownet to create optical flow for my dataset triplets. My question is how do I incorporate these flows into the train.py file, and which block of code do I need to modify to achieve this task.

Thank you, and I hope to hear from you soon.

hzwer commented 2 years ago

Hi, in our current version, we do not need to generate flow label for training. If you need this supervise, I advise add flownet to model.py add modify the loss_distillation.