Open marlluslustosa opened 4 years ago
Funny you brought this up - I actually had this on my list to implement :) Haven't had the time but hopefully soon. The Colab that you mentioned works for fixed frames 10 and 11 and it's just a matter of generalizing it into a loop that takes batches of frames at a time and interpolates it. That's my theory at least!
I'm sorry I didn't respond to you so soon. With the danger of the corona virus I have been working 5x more than normal and everything home office.
Well, I'm glad you had that in mind (to add DAIN). Who knows when to pre release a preview I can help you with the code.
Thanks again for the answer.
Hello, first I would like to congratulate you for your initiative.
I found your project by researching the EDVR git (https://github.com/baowenbo/DAIN/issues/44#issuecomment-583986332)
I have an incomplete job combining DAIN (interpolate at 60fps) and SuperSlomo (script to break the video into several images and then generate it again). I also thought of a shell script using ffmpeg to pre-process the video, turning it into an image dataset.
When I met your project I saw that you are already more advanced than mine. I would like to know the possibility of merging the DAIN with your code, after processing images. It would basically be an interactive option from EDVR or DAIN, just like you have done so far. Specifically on the last screen (Form Cell with a few options)
I am also very interested in popularizing Deep Learning for the general public who want to use pre-trained models in their projects (do-it-yourself style).
Congratulations one more time. :P