Study, Design, and Implement the Knowledge-aware Object Detection Symbolic Knowledge Injection Algorithm

Object detection is one of the most popular computer vision tasks, with solutions to the problem existing since before the spread of modern convolutional neural networks, or the newer transformer architecture. Applications can be found in a wide range of fields, from autonomous vehicles and surveillance, to medical imagining and wildlife tracking. The current project aims to implement and, potentially, improve a framework presented in a 2017 paper [FKL+17]. This framework is based on the notion of ”semantic consistency”, adding background information about the real world, and works by including external knowledge from a knowledge graph, alongside the image information. It reports an increase of 6.3 points in recall, compared to the bare-bone object detection solutions.