Welcome to the "Object Detection for Pedestrian Recognition" project repository! 🚶♂️🔍 This exciting endeavor delves into the world of modern computer vision algorithms to tackle the complex task of pedestrian recognition. Leveraging the power of transfer learning and cutting-edge techniques, we explore the effectiveness of different algorithms in identifying pedestrians amidst various challenges.
In this project, we roll up our sleeves to harness the potential of transfer learning in object detection. 🌟 Through the lens of three distinct algorithms—Single Shot MultiBox Detector (SSD300) with VGG16 backbone, Faster R-CNN with ResNet50 backbone, and Faster R-CNN with MobileNet backbone—we set out to crack the code of pedestrian recognition.
Our trusty companion on this journey is the renowned Penn-Fudan Pedestrian dataset, a benchmark for pedestrian detection tasks. 🏃♀️💼
Get ready to dive into the world of computer vision, where we explore the nuances of various algorithmic strategies, dissect their strengths, and uncover their limitations. Through quantitative metrics like Validation and Training Loss, IOU Score, and Inference Time, as well as captivating visualizations, we paint a vivid picture of each algorithm's prowess.
This dissertation is a beacon for enthusiasts, scholars, and practitioners alike. It's not just about solving challenges today; it's about providing a springboard for enhancing object recognition systems of the future. 🚀🔬
Pre-trained model links - https://drive.google.com/drive/folders/1PMxvUX2NZmaBhpeyEBiBWAQ_Hzo1RwDR?usp=sharing