Search papers, labs, and topics across Lattice.
Singapore University of Technology
1
0
3
A novel decoupled autoencoder design unlocks 32x spatial compression for diffusion models while simultaneously boosting reconstruction fidelity and accelerating training convergence.
