search

alexlee-gk / video_prediction

Stochastic Adversarial Video Prediction
https://alexlee-gk.github.io/video_prediction/
MIT License
303 stars 65 forks source link

About sequence length #17

Closed happyday521 closed 5 years ago

happyday521 commented 5 years ago

Hi! I train my model with context_frames=10 and sequence_length=30, in other words, aiming to predict the next 20 frames. (Because of my Hardware limitation). I wonder whether I can generate the next 30 frames(rather than 20) when I test the model. Can I set the context_frames=10 and sequence_length=40 directly when testing? Or need I modify anything? Thanks!

alexlee-gk commented 5 years ago

Yes, you can do that. The test sequence length doesn't have to be the same as the training sequence length. If you are evaluating on the full-reference metrics (i.e. the metrics that need a ground-truth sequence), then you are limited by the length of the ground-truth videos. If you are just generating frames, those predictions can be arbitrarily long.

happyday521 commented 5 years ago

Thanks！Closed.