Semantic Segmentation with ResNet-50, ResNet-101, and MobileNet v3 — DeepLab v3 is a semantic segmentation model that can use ResNet-50, ResNet-101 and MobileNet-V3 backbones. This hands-on article explains how to use DeepLab v3 with PyTorch. 1. A Quick Introduction to Semantic Segmentation Semantic segmentation divides an image into semantically different parts, such as roads, cars, buildings, the sky, etc. Below is an example of an image…

In 2016, Tsung-Yi Lin et al. published a paper on Feature Pyramid Network (FPN). The team includes Ross Girshick and Kaiming He, well-known for the work of Faster R-CNN. They developed FPN as a multi-scale feature extractor. Then, they used it with Faster R-CNN, significantly improving detection accuracy. …

Better, Faster, Stronger — Joseph Redmon and Ali Farhadi made various improvements to YOLO v1. The resulted model YOLO v2 became better (more accurate) and faster than Faster R-CNN and SSD. They trained a stronger model (YOLO9000) with a new method, making it capable of detecting over 9000 object categories. So, the paper title…

Why Is SSD Faster Than YOLO v1 and More Accurate than Faster R-CNN? — In 2015, Wei Liu et al. published a paper on SSD (Single Shot MultiBox Detector). It was faster than YOLO v1 (the first version) and more accurate than Faster R-CNN. This article explains what made SSD so good. 1. Background 1.1. Motivation The first author Wei Liu was a University of North Carolina student…

You Look Only Once (The 1st Version) — YOLO is an object detection model that is very fast, pretty accurate, yet simple. This article explains the architecture of YOLO (the 1st version). 1. Background In 2015, Joseph Redmon (University of Washington) developed YOLO. One of his co-authors, Ross Girshick (Microsoft Research), published a paper for Faster R-CNN around the same…

Real-Time Object Detection with RPN — There was no doubt Fast R-CNN was faster than R-CNN. What used to take 47 seconds per image went down to 0.22 seconds. That is, Fast R-CNN was 213 times faster than R-CNN. It significantly improved inference speed, albeit excluding the region proposal step that took around 2 seconds. It…

Understanding why it’s 213 Times Faster than R-CNN and More Accurate — In 2013, Ross Girshick et al. introduced R-CNN, an object detection model that combines convolutional layers with existing computer vision techniques, breaking previous records. It was a groundbreaking model at the time. In 2015, Ross Girshick developed Fast R-CNN, setting a new record. It was more accurate, and the inference…

R-CNN = CNN Extracting Features + SVM Classifier — R-CNN (Region-based Convolutional Neural Network) was an epoch-making model in 2013 which successfully combined CNN with classical computer vision techniques for object detection and broke the previous record. R-CNN is now an old model, but it’s essential to have knowledge of the origin in studying the subsequent development in object…

Understanding the Logic and Tricky Part of NMS Object detection models like YOLOv5 and SSD predict objects’ locations by generating bounding boxes (shown in blue rectangles below). However, object detection models produce more bounding boxes than the final output with different locations, sizes, and confidence levels. They do not just…