Introduction

(updated 2024.2).

Image Processing with Deep Learning

Deep learning innovations are driving breakthroughs in the field of computer vision for automotive, robotics, and science.

Thus, the aim of this course is to introduce students of computer vision starting from basics of digital image processing and then turning to modern deep learning models, specifically convolutional neural networks(CNN), for image enhancement and analysis applications on Energy/Environment, Autonomous Vehicle/Robots, and Machine Vision.

Students will learn how to build CNN(Convolutional Neural Network) based object detection model for vehicle, pedestrians and other target objects. Also, they will learn how to apply pre-processing and post-processing for CNN model input/output, such as calibration, filtering, image scaling, image augmentation using fundamental methods of computer vision.

This course is composed of two parts:

(1) Part1 : Digital Image Processing (8 weeks) Basics of digital image processing for pre/post-processing of CNN models will be covered including cameral calibration, spatial filtering, feature recognition, color image processing, image scaling/rotating and more. Students will learn how to apply image processing methods to build their own defect inspection and road lane detection system.

(2) Part 2: Convolutional Neural network for Object Detection (8 weeks) Basics and useful technique in building a CNN model will be covered including backpropagation, evaluation metric, regularization and more. Also, the state-of-the-art CNN models such as Inception and Yolo v3 will be introduced. Students will learn how to use their own customized dataset to train and test a CNN object detection model.

After taking this course, students will be able to build their own program for

  • Road lane detection

  • Vehicle/Pedestrian detection

  • Face recognition

  • Defect Inspection

  • and more

Lecture Syllabus

Part 1

  • Introduction to Image Processing

  • Optics / Calibration

  • Tutorial: OpenCV, Calibration

  • Spatial Filtering

  • Histogram/Threshold

  • Morphology

  • Edge,Line,Corner Detection

  • Color Image Processing

Part 2

  • Inro. Deep Neural Network: MLP, Activation Function, Loss Function

  • Inro. Deep Neural Network: Back Propagation, CaseStudy(LeNet)

  • Intro. Convolutional Neural Network: Convolution, Evaluation Metric

  • Intro. Convolutional Neural Network:Strategy for CNN design, Case Study(AlexNet)

  • Popular CNN models: VGGNet, ResNet, Inception

  • Object Detection: from R-CNN to YOLO

  • Recent trend in CNN

Tutorial

  • Tutorial: OpenCV, Calibration

  • Tutorial: Image Spatial Filtering

  • Tutorial: Morphology

  • Tutorial: Edge & Line Detection

  • Tutorial: Color Image Processing

  • Tutorial: PyTorch

  • Tutorial: LeNet

  • Tutorial: AlexNet

  • Tutorial: AlexNet with Customized Dataset

  • Tutorial: YOLOv8

LAB

  • LAB: Object Segmentation in Grayscale Image

  • LAB: Object Segmentation in Color Image

  • LAB: Industrial Problem

  • LAB: CNN Object Detection (Vehicle /Pedestrian)

  • LAB: CNN Object Detection (Custom dataset, Free Topic)

Last updated