Releases: lucidrains/video-diffusion-pytorch
Releases · lucidrains/video-diffusion-pytorch
0.3.10
make sure initial temporal attention gets rotary embeddings
0.3.9
bring in an initial temporal attention after the first initial convol… …ution, to see if it can help with the flickering
0.3.8
remove convnext blocks, as seeing first good results on moving mnist,… … thanks to co-debugging with @mrkulk
0.3.7
experiment tracker agnostic
0.3.6
change focusing on present to be a probability, so that a single batc… …h could contain certain percentage of videos that have arrested attention across time
0.3.5
start using torch inference mode
0.3.4
add additional rotary positional embedding for time attention
0.3.3
allow for gradient clipping
0.3.2
allow for gradient clipping
0.3.1
make sure some hyperparameters for resnet and convnext blocks are con… …figurable