Computer Vision with Python

train a binary classifier to detect whether a handwritten digit is five using an svm, then extend to a multiclass model and evaluate with a confusion matrix.

Learn to train a multi-class classifier by importing the classifier, initializing it, fitting on the training data, and predicting on the first input.

Evaluate the confusion matrix after training a binary classifier by predicting on the train data and computing true positives, true negatives, false positives, and false negatives.

Learn Prewitt edge detection by applying horizontal and vertical 3x3 kernels via convolution to build edge maps, with prior gaussian noise reduction using cv2 and filter2D.

Explore Sobel edge detection for stable edges by applying horizontal and vertical filters, following the same procedure as the previous edge detection.

Apply a filter to approximate second-order derivatives for Laplacian edge detection and compute the image's edges, with noise reduction beforehand since the method is highly sensitive to noise.

Explore convolutional neural networks used in image and speech recognition, learn about convolutional layers and fully connected layers, and implement CNNs in Python with exhaustive search.

Understand how convolutional layers apply a convolution over inputs with a receptive field to produce feature maps, then use max pooling and a fully connected layer in a simple architecture.

Explore famous CNN architectures such as visual net and inception modules, featuring conv layers, pooling, fully connected layers, and soft max, and learn about residual networks with 152 layers.

Apply a cnn python implementation pattern to load and preprocess images, load a pretrained model, predict class probabilities, and decode top predictions with examples like tiger, cat, and flower.

Explore exhaustive search for object localization by sliding windows of varying sizes across images, computing multiple locations to enable object detection, despite higher computation.

Learn how to get data, read it, and build and train a convolutional neural network for classification using Kaggle and a tiny 200-class ImageNet dataset.

Read data by building paths to training and validation images, then load images with cv2; map class IDs to indices and assemble training and test arrays from the annotation file.

Build a ten-class inception-based model with global average pooling and a dense output, then train with SGD to reach 97% accuracy, comparing to an Alexa net style model at 85%.

Set up a cloud training environment on a digital ocean droplet, transfer data and scripts, install dependencies, and run training that reaches 97% accuracy after five epochs on image net.

Celebrate completing the course and explore optional one-to-one sessions for personalized guidance, live troubleshooting, and real-world application q&a, bookable by emailing Lucas at 50794 at gmail.com.