Love the paper / model, just wanted to let you guys know you can (slightly) improve the fid even further by using the learned timestep uncertainty / logvar to add an appropriate amount of noise during the sampling / inference process at each timestep.
Hi guys,
Love the paper / model, just wanted to let you guys know you can (slightly) improve the fid even further by using the learned timestep uncertainty / logvar to add an appropriate amount of noise during the sampling / inference process at each timestep.