Real time object detection in video using YOLO on iPhone

Discover how deep learning enables machines to learn patterns from labeled images, automatically extract features across hierarchical layers, and perform real-time object recognition in scenes.

Explore the differences between artificial intelligence, machine learning, and deep learning, and how ml uses labeled data while dl learns features automatically with neural networks.

Experience why deep learning is taking off through rapid hardware advances, parallel gpu computation, and exploding data availability that boost neural network performance beyond traditional machine learning.

Demonstrates using deep learning to model real estate prices with neural networks, activation functions like ReLU, and multiple features such as size, bedrooms, and ZIP code.

Explain logistic regression as a binary classifier that uses input image pixels and weights to predict cat or dog, and demonstrate weight updates via gradient descent to minimize loss.

Explore how the cost function in logistic regression measures prediction error with a sigmoid output and guides optimization via gradient descent on W and B across training samples.

Explore gradient descent and back propagation for neural networks, learning to minimize loss via forward and backward propagation, computation graph, and updates to weights and biases with learning rate alpha.

Examine the multilayer perceptron, from perceptron blocks and activation functions like sigmoid and relu to forward and backpropagation, with applications in sentiment analysis and real estate price prediction.

Explore how convolutional neural networks extract features from images, using convolution, pooling, relu activation, and softmax for real-time object detection.

Describe how yolo uses a seven by seven grid where each cell predicts bounding boxes and objectness, then combines class probabilities with non maximum suppression to output final detections.

Set up a live back-camera capture session and stream frames in real time, processing each frame with a model to detect objects, draw bounding boxes, and show names with confidence.

Set up a new Xcode project as a single view app, organize the folder, and add a camera usage description in Info.plist for object detection.

Set up live capture initializes a back camera feed, creates an AVCaptureSession, and configures input and video data output for real-time object detection using a vision-based machine learning model.

Connect the camera preview layer to the user interface by configuring a preview layer, setting video gravity, and attaching it to the view; then start the capture session.

Map device orientation to cg image orientation using an exif-based function to ensure the video frames align for the real-time yolo model, handling landscape, portrait, and upside-down modes.

Set up the detection layer that overlays red-bordered rectangles on the camera feed, with a transparent yellow background and overlaid text, to render object detections from the model.

Import a YOLOv3 model into the app, set up a model inference request, process video frames frame-by-frame, and draw bounding boxes with labels for real-time object detection.

Draw real-time bounding boxes around detected objects in video on iPhone, using model observations to generate labels, coordinates, and the confidence rate with dynamic UI updates.

This lecture explains debugging and implementing real-time YOLO v3 bounding boxes on iPhone, including shape layers, color properties, and overlaying labels and scores on camera frames.

Demonstrates running a real-time object detection app on iPhone and iPad, streaming live video, detecting people and objects with bounding boxes and confidence scores in real time.

Discover next steps in real-time object detection on iPhone using YOLO, exploring CNN basics, bounding boxes, confidence thresholds, and applying to your own videos and images.