End to end object detection using transformer neural networks

Abstract

Object detection plays a crucial role in computer vision, identifying specific object newlineclasses like humans, animals, and vehicles in images or video frames. The main objective newlineis to create computational model that addresses the fundamental task of locating the newlineposition of objects in an image. This task forms the basis for various downstream computer newlinevision tasks such as instance segmentation, image captioning, and object tracking. newlineIt encompasses a wide range of applications like detecting pedestrians, animals, vehicles, newlinepeople, faces, text, and more. Recent years have seen rapid breakthroughs in newlineobject detection due to advancements in deep learning techniques. The combination of newlinedeep learning networks and GPU processing power has led to significant breakthroughs, newlineimproving the efficiency of object detectors and trackers. newlineAttention mechanisms were introduced to enhance the performance of encoderdecoder newlinemodels in machine translation. These mechanisms empower the decoder to newlineflexibly utilize relevant segments of input sequences, giving more weight to regions or newlineaspects of the input image that are considered more informative for identifying objects. newlineIn 2020, transformers entered the realm of computer vision with the Vision Transformer newlinedesigned for image classification. These models are reshaping the computer vision landscape newlineby becoming the standard for fundamental tasks like classification and object detection. newlineThe transformer architecture has been leveraged to achieve cutting-edge results newlinein object detection. The success of transformer-based architectures in natural language newlineprocessing has spurred their adoption in computer vision. The thesis explores the potential newlineof attention mechanisms and transformer-based models for object detection, aiming newlineto enhance accuracy and efficiency compared to traditional convolutional neural networks newlineand delves into how attention mechanisms and transformers have transformed newlineobject detection. newlineThe main goal of this study is to create advanced models for recognizing objects, newlinespecifically