Single Shot Multibox Detector | SSD Object Detection Explained and Implemented

In this video, I get into Single Shot Multibox Detector or SSD, a popular real-time object detection model. We will understand how Single Shot Multibox Detector algorithm works, and also do step by step walkthrough of implementation of SSD in PyTorch. This video is part of my object detection series, where I’ve previously covered YOLO, and now we’re exploring SSD object detection to get an understanding of how it combines aspects of YOLO and RCNN with the idea of using multiple feature maps. We first start with a high level overview of SSD object detector, then understand each and every detail of default boxes. We understand how these default boxes are matched during training ssd for object detection as well as loss. We then get into ssd model architecture and finally cover entire implementation of single shot multibox detector in PyTorch as well as look at its results. With this video one should be able to have a clear understanding of how this model works and be able to train ssd detector on the task of real-time object detection on their custom dataset. And throughout the video we draw comparisons between ssd, yolo and rcnn not just in terms of results but also their methodology. ⏱️ Timestamps: 00:00 Intro 00:38 Single Shot Multibox Detector Intro 05:15 Default Boxes in SSD 12:16 Matching Strategy used in SSD 15:51 Loss 17:26 SSD Model Architecture 24:04 SSD Object Detector Implementation 25:38 Data Augmentation used in SSD training 30:13 SSD Model Implementation 38:47 SSD Default Boxes Implementaiton 42:23 TrainingLoss Implementation for SSD 48:08 Inference time Post-processing and transformation 50:42 Results and Experiments 🔔 Subscribe : https://tinyurl.com/exai-channel-link 📖 Resources: SSD Paper - https://tinyurl.com/exai-ssd-paper Github Implementation Link - https://tinyurl.com/exai-ssd-implementation Background Track - Fruits of Life by Jimena Contreras Email - explainingai.official@gmail.com