AI, Deep Learning and Computer Vision with Python BootCamp

Explore how deep learning automatically learns hierarchical representations from raw data for image analysis, surpassing traditional handcrafted features and enabling end-to-end learning and transfer learning across domains.

Learn to write and run Python code entirely in google colab, a no-setup hosted jupyter notebook with free GPUs/TPUs, browser-based sharing via google drive.

Build a convolutional neural network with Python and PyTorch, featuring two conv layers with ReLU, two max pooling layers, and three fully connected layers, plus forward flow and size reasoning.

Learn how to optimize deep learning models by tuning hyperparameters and selecting optimizers like Adam, using cross-entropy loss to minimize loss on CIFAR-10 image classification.

Validate the convolutional neural network on unseen test images with a no-grad context and tally correct predictions. Print the accuracy calculated as true positives plus true negatives over all predictions.

Evaluate cnn performance using accuracy, precision, recall, and F1 score with true positives, true negatives, false positives, and false negatives. Guide improvements by analyzing strengths and weaknesses with these metrics.

Visualize the confusion matrix and compute precision, recall, and F1 score with a full evaluation pipeline using sklearn, plotly, and seaborn, including per-class performance.

Construct a deeper convolutional neural network for image classification by adding four convolutional layers with two pooling layers and two fully connected layers, with configurable input/output channels and kernel sizes.

Apply data augmentation to CIFAR-10, resize images to 32 by 32, train a deeper CNN with learning rate decay and tuned batch size, and evaluate with confusion matrices.

Learn to perform single-label and multi-label image classification with pre-trained models in PyTorch, using transfer learning with ResNet and Lex Net trained on ImageNet.

Explore ResNet and AlexNet architectures for image classification, including 18-152 layer deep CNNs, their convolutional and pooling structures, and the use of pre-trained models on ImageNet for research.

Place data in Google Drive, create a data folder, and upload dog, cat, and pizza images. Read images into Colab with PIL and image.open, then display them with plt.imshow.

Classify images with pre-trained ResNet and AlexNet models from torchvision, comparing ResNet variants and mapping predicted ids to ImageNet class names for single-label classification.

Fine-tune a pre-trained ResNet 101 for your dataset by replacing its final fully connected layer, enabling gradient updates, and training on GPU through backpropagation.

Train a ResNet fine-tuning model using PyTorch with 20 epochs, save and load from Google Drive, and visualize results showing bread, coffee, and pasta predictions.

Perform transfer learning with a pre-trained resnet 101 as a fixed feature extractor. Freeze all parameters and train only the last fully connected layer on the new dataset.

Learn how object detection combines classification and localization to identify objects and draw bounding boxes with class labels. Explore region proposals like selective search, edge boxes, and RPNs.

Explore object detection with pre-trained Detectron2 models, load configurations from the model zoo, and compare faster r-cnn with fpn against retinanet on the Coco dataset, visualizing predictions.

Configure Google Colab GPU and mount Google Drive to access the balloon dataset; convert it to Detectron2 COCO format, register it, train, and visualize annotations.

Explore how YOLO detects objects in a single pass by predicting bounding boxes, confidence scores, and class probabilities on a grid with anchor boxes, followed by non-maximum suppression.

Explore YOLO eight, the Ultralytics model featuring anchor-free detection, the C2F backbone, and a decoupled head, with mosaic augmentation and advanced losses for robust object detection, segmentation, and pose estimation.

Train a YOLO version eight model on the vehicle detection dataset with train/val/test images for five classes, configuring dataset YAML paths, then perform video object detection.

Train the YOLOv8 small model on the vehicles dataset (ambulance, bus, car, motorcycle, truck) for 30 epochs with 416 by 416 images and batch five.

Run inference with the YOLO eight model on test images and videos using weights and a confidence threshold to produce predictions saved in runs/detect, detecting ambulances, cars, buses, and motorcycles.

Learn to evaluate object detection models using intersection over union, average precision, mean average precision, precision, recall, and F1 score, and to validate these metrics on the validation dataset.

Export your trained YOLO8 model to multiple formats with Ultralytics YOLO8 export mode, enabling deployment across platforms using PyTorch, TorchScript, CoreML, tf lite, TensorRT, OpenVINO, and more.