Search papers, labs, and topics across Lattice.
2
0
4
Discriminator-Guided RL achieves a remarkable reduction in FID from 9.38 to 2.62, showcasing a new way to align model outputs with real data without human preferences.
Text-to-image models get a diversity boost without sacrificing quality, thanks to a new sampling method that disentangles and controls prompt-dependent and independent variations in the CLIP embedding space.
