Pre-training with Random Orthogonal Projection Image Modeling

arXiv:2310.18737v2 Announce Type: replace-cross
Abstract: Masked Image Modeling (MIM) is a powerful self-supervised strategy for visual pre-training without the use of labels. MIM applies random crops to input images, processes them with an encoder, and then recovers the masked inputs with a decoder, which encourages the network to capture and learn structural information about objects and scenes. The intermediate feature representations obtained from MIM are suitable for fine-tuning on downstream tasks. In this paper, we propose an Image Modeling framework …

abstract arxiv crops cs.ai cs.cv cs.lg decoder encoder image images information inputs labels learn modeling network pre-training processes projection random strategy them training type visual