Gain insights into automated inspection using computer vision. Learn image analysis, feature detection, 2D/3D transformations, and train neural networks for object detection and image segmentation.

Automated Inspection with Computer Vision.png

AICV_jupyterOpenCV.tar.gz

load_image

annotation

color_spaces

thresholding

smoothing

masking

jup_smooth

morphology

blob_detection

jup_fiducials

edge_detection

jupyter_masking_edge_detection

corner_detection

jupyter_corner_detection

homographies

jup_homographies

topography

jup_topography

tensor_manipulation

neural_network_building_blocks

training_neural_network

jup_train_nn

cnn_building_blocks

cnn_for_classification

pretrained_classification_cnn

jup_train_classification_cnn

transfer_learning

jup_finetune_classification_CNN

cnn_for_object_detection

jup_pretrained_object_detection

cnn_for_semantic_segmentation

jup_pretrained_semantic_segmentation

annotation_formats

spa_train_mlp

prj_jup_create_train_mlp_classification

Computer vision is essential for developers who wish to learn practical skills, e.g., in industrial manufacturing, automated inspection of products is crucial for quality assurance.

In this course, you’ll apply computer vision and machine learning to analyze images for automated inspection. You’ll start by learning image I/O operations, thresholding, smoothing, and masking. You’ll learn feature detection using template matching and morphology. You’ll use the Sobel and Canny Edge Detectors, Harris Corner Detector, and Hough transform. You’ll learn about 2D transformations, including perspective and affine transformation, and 3D-to-2D projections. You’ll perform topography with a laser line to inspect the 3D shape of an object. Lastly, you’ll train convolutional neural networks for object detection and image segmentation and annotate datasets using CVAT.

By the end of this course, you will have the skills to build inspection systems for industrial applications using computer vision and machine learning.

Automated Inspection with Computer Vision

**(True or False)** Semantic segmentation CNNs return a bounding box for each found object.

(True or False) Semantic segmentation CNNs return a bounding box for each found object.

**(True or False)** Multiple objects of the same category that touch each other would be separated with distinct blobs by a semantic segmentation CNN.

(True or False) Multiple objects of the same category that touch each other would be separated with distinct blobs by a semantic segmentation CNN.

An image batch tensor of shape `(2, 3, 520, 600)` is passed through a semantic segmentation CNN that was trained to detect objects from 21 categories. What is the shape of the ```out``` tensor, from the returned dictionary?

An image batch tensor of shape <code>(2, 3, 520, 600)</code> is passed through a semantic segmentation CNN that was trained to detect objects from 21 categories. What is the shape of the <code>out</code> tensor, from the returned dictionary?

We get the color at pixel `(x, y)`, and find the mapping between the color and the category index.

We get the index of the only channel having a non-zero value, at pixel `(x, y)`.

It is the index of the channel whose value is highest at pixel `(x, y)`.

Assume that the semantic segmentation for an image is a tensor of shape `(21, 520, 600)`. How do we get the category index associated with a given pixel `(x, y)`?

Assume that the semantic segmentation for an image is a tensor of shape <code>(21, 520, 600)</code>. How do we get the category index associated with a given pixel <code>(x, y)</code>?

**(True or False)** For a given pixel, only the class with the highest confidence level is available from the semantic segmentation tensor.

(True or False) For a given pixel, only the class with the highest confidence level is available from the semantic segmentation tensor.

**(True or False)** When resizing an image that annotates the highest confidence class with a color code, the interpolation method matters.

(True or False) When resizing an image that annotates the highest confidence class with a color code, the interpolation method matters.

If we can perform object detection, we get the semantic segmentation of the object for free.

The number of object categories in semantic segmentation is limited to 64.

If we can perform semantic segmentation of an object, we get its object detection for free.

The dictionary that maps the object category to a color is arbitrary.

**(Select all that apply.)** What is _true_ about semantic segmentation?

(Select all that apply.) What is true about semantic segmentation?

Semantic Segmentation CNNs

Semantic Segmentation CNNs

Test your knowledge on semantic segmentation CNNs.

Introduction

Getting Started with Images

Image I/O and Annotations

Color Spaces and Thresholding

Convert Color Spaces, Threshold

Smoothing and Masking

Detection of Features

Image Registration

3D Vision

Getting Started with Neural Networks

Convolutional Neural Networks

Project: Create and Train a CNN for Classification

Object Detection and Semantic Segmentation

Cats vs Dogs Classification with Convolutional Neural Networks

Dataset Annotation

Final Remarks

Recognize Handwritten Digits Using a Deep Neural Network

Quiz: Semantic Segmentation CNNs