Article-Journal

DNNs May Determine Major Properties of Their Outputs Early, with Timing Possibly Driven by Bias

SeiT++: Masked Token Modeling Improves Storage-efficient Training

Probabilistic Language-Image Pre-Training

Few-shot font generation with weakly supervised localized representations

Styleaugment: Learning texture de-biased representations by style augmentation without pre-defined textures

Semantic-aware neural style transfer