SD-DiT: Unleashing the Power of Self-supervised Discrimination in Diffusion Transformer

arXiv:2403.17004v1 Announce Type: new
Abstract: Diffusion Transformer (DiT) has emerged as the new trend of generative diffusion models on image generation. In view of extremely slow convergence in typical DiT, recent breakthroughs have been driven by mask strategy that significantly improves the training efficiency of DiT with additional intra-image contextual learning. Despite this progress, mask strategy still suffers from two inherent limitations: (a) training-inference discrepancy and (b) fuzzy relations between mask reconstruction & generative diffusion process, resulting in sub-optimal training …

abstract arxiv convergence cs.cv cs.mm diffusion diffusion models discrimination efficiency generative image image generation power strategy training transformer trend type view