Billion-Scale Pretraining with Vision Transformers for Multi-Task Visual Representations | IEEE Conference Publication | IEEE Xplore