Data mixing augmentation has proved effective in training deep models. Recent methods mix labels mainly according to the mixture proportion of image pixels. Due major discriminative information a fine-grained usually resides subtle regions, these tend introduce heavy label noise recognition. We propose Semantically Proportional Mixing (SnapMix) that exploits class activation map (CAM) lessen au...

In asymmetric stereoscopic video compression, the views are coded with different qualities. According to the binocular suppression theory, the perceived quality is closer to that of the higher-fidelity view. Hence, a higher compression ratio is potentially achieved through asymmetric coding. Furthermore, when mixed-resolution coding is applied, the complexity of the coding and decoding is reduc...

We give various examples of asymmetric orbifold models to possess intertwining currents which convert untwisted string states to twisted ones, and vice versa, and see that such asymmetric orbifold models are severely restricted. The existence of the intertwining currents leads to the enhancement of symmetries in asymmetric orbifold models. ∗ To appear in the proceedings of YITP Workshop on “Rec...