MaxVit -- Multi Axis Vision Transformer

Over the past few years, there has been a lot of competition of iterative improvements based on Convolution Nets and the relatively recent Transformer in terms of being the best architecture on the standard Image vision tasks. In the paper published in ECCV 2022, Researchers in Google Research and UT Austin introduce MaxVit. MaxVit -- Multi Axis Vision Transformer aims to combine the best features of both Convolution and Transformer by solving the issue related to global attention in transformers. We will first discuss the set of Vision Task for which these methods are applied .A typical vision task involved taking the input of 2D image and taking the RGB matrix format to your Neural network architecture. Image Classification is the problem of assigning labels to images from a fixed set of categories.