[NEW] 2026:Build 15+ Real-Time Computer Vision Projects

Classify handwritten digits with the mNIST dataset by building a model using the Keras sequential or functional API. Save, reload, and predict from the trained model.

Learn to build convolutional neural network image classifiers for MNIST using both sequential and functional APIs, including data preprocessing, 28x28x1 inputs, and training with Adam and cross entropy.

Classify fashion MNIST images with a CNN using Keras. Train with early stopping, save and reload the model, then make predictions across classes like t-shirt, dress, coat, and sneaker.

Learn to build a fashion MNIST image classifier with a CNN, including data reshaping, one-hot encoding, convolutional layers, training with categorical cross-entropy, and saving the model for deployment.

Explore the image transformation project by applying random crops, flips, translations, rotations, zooms, and contrast adjustments to dog and cat images using Keras preprocessing layers.

apply keras preprocessing layers to augment cats and dogs images from a directory in Colab, using resize, random flip, crop, translation, rotation, brightness, and contrast.

Apply transfer learning to build x-ray classifier using transformations like random crop and flip; generate images with image generator, and fine-tune a custom pre-trained model to distinguish normal from pneumonia.

Combine pre-trained models (MobileNet v2 and DenseNet 169) via a concatenation in a functional API to enable transfer learning for chest x-ray classification (normal vs pneumonia) with data augmentation.

Use generative adversarial networks to generate image captions by combining provided images with their descriptions. Explore practical GAN-based image captioning with a dataset of images and descriptions.

Build an image captioning pipeline by encoding images with Inception v3 features and using glove embeddings. Generate captions with an lstm using start sequence and end sequence, and prepare dataset.

Explore a GAN-based image captioning solution, including image encoding, caption cleaning, vocabulary construction with GloVe embeddings, and an LSTM-based data generator.

this lecture explains building a two-input image captioning model that fuses encoded image features with caption sequences, using embedding, dropout, an lstm decoder, and categorical cross entropy training.

Learn to annotate images for computer vision projects using tools like Make Sense and Roboflow, creating labels, bounding boxes, and polygons, and exporting annotations in COCO, VOC, or YOLO formats.

Explore object detection with YOLOv5 on a hard hat versus safety vest dataset, training models to classify and detect hard hats, helmets, and safety vests in images and videos.

Learn to build an object detection system with YOLOv5, train on a custom hard hat and safety vest dataset, and deploy predictions using the Ultralytics YOLOv5 workflow.

Classify weather scenes in images and videos using YOLOv8-cls by training on a weather dataset to predict sunny, cloudy, windy, snowy, or rainy conditions.

Organize a weather image dataset into train, test, and validation folders and perform a randomized split. Train a YOLOv8-cls classifier, evaluate results, and save the trained weights.

Carry out instance segmentation on a potholes dataset by annotating the images and generating separate masks for each pothole, using YOLOv8-seg to create precise segmentation.

Apply instance segmentation with YOLOv8-seg to pothole images by configuring Ultralytics, loading data from Google Drive, training or fine-tuning a segmentation model, and generating pothole masks and visual results.