deepmind-research
deepmind-research copied to clipboard
Perceiver IO: Video Autoencoding notebook - reconstruction of full video doesn't display properly or is broken
Running on Google Colab GPU the Visualize reconstruction of first 16 frames display correctly, but not the reconstruction of full video displays like this and the sound doesn't play:
Thanks for reporting this issue! This should now be fixed.
Hi, thank you looking into it :-) . The video at the bottom now correctly displays but only shows the first segment (the same 16 frames as above). Changing the index from [0] to [n] does show the subsequent segments, but I expect that you intend to play the full video in one go.
Hi! Just bumped into the same issue with this notebook, and fixed the last cell by rearranging the image output with einops:
from einops import rearrange
full_video = rearrange(reconstruction['image'], 'batch frames w h c -> (batch frames) w h c')
table([to_gif(full_video), play_audio(np.array(reconstruction['audio'][0]))])
Hope this helps! ☺️