Machine Learning: Modern Computer Vision & Generative AI

Explore image classification with convolutional neural networks, learning to assign discrete labels to images using a pre-trained classifier and fine-tuning with transfer learning.

Explore pre-trained image classifiers, often trained on imageNet, to save time and improve performance on new images. Understand backbone and heads, transfer learning, and limitations like 1000 classes and generalization.

Learn to run a pre-trained image classifier with Keras CV in Python, load EfficientNet v2 b0, preprocess images to a batch, and map predictions to human class labels.

Leverage transfer learning and fine tuning to adapt pre-trained CNNs and transformers to new tasks, using pre-training on large datasets and targeted final layer updates with limited data.

Fine-tune an image classifier in Python using Keras on your own lions and cheetahs dataset in Colab with a GPU, loading data via tf.keras.utils.image_dataset_from_directory and a 0.3 validation split.

Choose or build a labeled image dataset, organize images by class, and run classification code to compare accuracy and backbone performance, noting speed–accuracy trade-offs.

Submit feedback through the suggestion box to help tailor this course on modern computer vision and generative AI. Share your background, difficulty, and topics like CNNs, transformers, and gradient boosting.

Explore object detection fundamentals, decode model predictions, and train detectors with Keras CV, IOU, and non-max suppression, using Coco and Pascal VOC data, and fine-tuning on pre-trained or custom datasets.

Explore how object detection extends image classification to locate multiple objects with bounding boxes, using annotations and regression, and loss components like binary cross entropy and categorical cross entropy.

Decode object detection outputs by mapping fixed predictions to anchor boxes, applying the confidence score and IOU-based non-max suppression to produce final bounding boxes and class predictions.

Explore pre-trained object detection with RetinaNet ResNet-50 Pascal VOC in a Python notebook. Load the model, prepare images, run predictions, plot bounding boxes, and tune non-max suppression to improve detection.

Explore smooth L1 loss, which blends squared and linear behavior to stabilize training, and focal loss, which uses alpha and gamma to address class imbalance in object detection.

Explore COCO and Pascal VOC object-detection data formats, including JSON structures with categories, images, and annotations, and XML per image, bounding-box coordinates, and conversion considerations for Keras CV.

Learn how to set up LabelImg, a gooey program for annotating images for object detection with a mouse, using Anaconda, and run the gui with label img.

Demonstrates installing labelImg on mac with Anaconda, annotating images using the GUI, and exporting annotations to Pascal VOC XML and COCO JSON formats.

Apply data augmentation techniques such as rotation, scaling, and cropping to boost training data for computer vision and object detection, preserve ground-truth labels, and improve generalization while reducing annotation effort.

Learn to convert coco format object detections into a tf dataset for keras cv, building image, classes, and bounding boxes as ragged tensors and dict-to-tuple formats.

Train object detection from scratch using Keras, TFDS, and RetinaNet on Pascal VOC data; manage bounding box formats, data augmentation, and model training with focal loss.

Fine-tune an object detection model in Python on a custom dataset using RetinaNet with a ResNet-50 backbone, covering data loading, class mapping from COCO, bounding boxes, augmentation, and visualization.

Practice what you learned by building a custom object detection dataset from scratch, training a detector, and tuning IoU and confidence thresholds to maximize mean average precision.

Explore generating images with stable diffusion through hands-on code in the Keras CV library, then optional high-level explanations. Compare open models such as Stability AI's diffusion, Dall-E, and Midjourney.

Learn to generate images with stable diffusion in python using keras, covering model setup, text-to-image prompts, and performance tricks like mixed precision, jit compilation, across three notebooks.

Diffusion models generate high quality images and sound by reversing a forward noising process, and the talk surveys autoencoders, variational autoencoders, and GANs.

Explore diffusion model architectures that predict image noise with a U-Net and skip connections, trained self-supervised on noisy inputs, and latent diffusion used in stable diffusion.

Condition diffusion models on prompts by injecting text embeddings into every layer of the u-net. Understand how transformer blocks and position embeddings drive prompt adherence via cross attention.

Explore how the Keras diffusion model implements stable diffusion through a latent diffusion U-net with text and time-step embeddings, spatial transformers, and cross-attention.

Clarify the appendix as the FAQ, a set of optional, commonly asked questions with supplementary material, and encourage using the Q&A for detailed answers.

Perform a pre-installation check with guidelines for Python and libraries, emphasize understanding concepts over syntax, and note when to install libraries like OpenAI Gym in relevant courses.

Beginner's coding tips lecture explains why coding along can mislead beginners, stresses thinking speed over typing, and offers self-guided coding resources and references like the lazy programmer materials.

Code by yourself to learn machine learning, embracing a practice loop that translates pseudocode into code using a consistent fit and predict pattern for supervised learning.

Embrace test driven development by writing tests first to shape your API and design, then implement theory in code, reinforcing intuition through practice.

Show that Jupyter notebook offers no real difference from plain Python code and emphasize writing and debugging outside notebooks with print statements and IPython as a REPL.

Master machine learning through a balanced blend of Python coding, math and derivations, clarifying prerequisites, and bridging academic rigor with practical, real-world implementation.

Discover how to learn machine learning topics in a skill-building sequence using a dependency graph, from numpy and linear regression to deep learning, CNNs, and recurrent networks.

Navigate the course path by understanding how supervised, unsupervised, and reinforcement learning tracks relate, and how prerequisites guide the order, including word embeddings and RNNs.

Learn how to get discount coupons and free AI content via the newsletter on lazyprogrammer.me or deeplearningcourses.com, and explore VIP material releases and permanent access.