Paper ties diffusion training speed to residual-gate layout
A teacher-free radial objective on ImageNet 256 with SiT-B/2 improves FID over dispersive loss while widening gate participation.
2 stories, newest first · one card per story
A teacher-free radial objective on ImageNet 256 with SiT-B/2 improves FID over dispersive loss while widening gate participation.
Joint path-flow training with a stochastic regularizer improves ImageNet-256 FID without changing inference.
Today 1.6K readers · 2.5K views · 779 reads · all time 189K views