Deep Learning Neural Networks with TensorFlow

Build a TensorFlow face mask detection app using Keras in Python, gaining hands-on experience with real-time image and video classification, data analysis, and pattern detection for business solutions.

Set up a face mask detection project in TensorFlow by installing libraries, preparing a train/test dataset (from Kaggle), and using a pre-trained MobileNet v2 with Keras in a Jupyter notebook.

Load and preprocess train and test images for a two-class mask vs no-mask classification using a pretrained MobileNet v2 in TensorFlow.

Define the model architecture and fit the model for five epochs using validation data. Monitor training progress as loss and accuracy are shown, with accuracy increasing.

After five epochs, save the trained mask detection model locally, then load the saved model and build a function to predict new data.

Write a function that loads a saved TensorFlow model, processes an image with cv2 (resize and convert BGR to RGB), and predicts mask versus no mask.

This lecture introduces TensorFlow, a Python library for fast neural numerical computing that uses data flow graphs of ops and tensors, including sessions and a hello world example.

Install TensorFlow on Windows 64-bit, choosing CPU or GPU support; use native pip or Anaconda in a virtual environment, with a recommended pip installation.

Discover TensorFlow basic data types such as float, double, various integers, unsigned integers, string, and boolean, and learn to build and run a computational graph with constants and sessions.

Implement a simple linear model with w and b on x, use placeholders to compute loss, and minimize it with a gradient descent optimizer, updating variables until convergence.

Create a Python file to implement a linear model in TensorFlow, using w and b variables and a placeholder, compute a loss, and repeatedly update parameters to reduce it.

Explore optimization of a loss value in TensorFlow by using gradient descent, configuring an optimizer, running training loops, evaluating loss, and applying feed dictionary and replace dictionary techniques.

Implement the constructor variable in TensorFlow with tf.Variable, perform addition and multiplication, initialize and run a session to evaluate results, and demonstrate placeholders with numpy, reduce prod and reduce sum.

Learn to add numbers in TensorFlow with a tf session and constants, printing 32, generate 100 data points to explore fitting y = 0.1x + 0.3 with w and b.

Initialize tf variables w and b, build a simple linear model y = w*x + b, and minimize the mean squared values errors with a gradient descent optimizer.

Build an image captioning model with TensorFlow. Learn image and text pre-processing, sequence models, Bleu evaluation, and deployment with Streamlit and AWS EC2.

Import TensorFlow and Keras libraries, along with image and text preprocessing tools (load_img, image_to_array, preprocess_input, tokenizer, pad_sequences), and model components (dense, LSTM, embedding, add) for a hands-on deep learning project.

Download the flickr8k caption dataset from Dropbox, unzip it, and review the text file of image IDs with multiple captions, then set up train and test caption paths for training.

Download the Flickr 8-K image data set from Dropbox, unzip it, and remove the zip file, then create all_images as the complete folder path to the images for training.

Preprocess text data from the Flickr dataset using Python to read caption files, parse image id and caption pairs, and add start and end tokens to each caption.

Pre-process captions by lowercasing and removing special characters with regular expressions, organize captions by image ID, build a vocabulary, and load captions from train and test image files.

Load image captions for training and test data using load_text, build Python dictionaries mapping image IDs to captions, and perform pre-processing on Flickr 8-K text.

Preprocess image data by listing files from all_images, load and resize to 224×224, convert to array, and extract features with a pre-trained VGG16, saving to features.dump.

Load features from a pickle file using VGG 16 and map them to train and test image IDs, forming a dictionary of 1000 features per image.

Learn how to prepare text data for an LSTM model by tokenizing sequences, building a tokenizer, computing vocab size and max caption length, applying padding, and creating input-output sequences.

Create a custom data generator to feed image features and caption sequences into a two input LSTM model, training with categorical cross entropy and Adam to predict the next word.

Define the model with vocabulary size and max length in TensorFlow, then train using a custom data generator and tokenizer for ten epochs, and save after each epoch.

Complete ten epochs of a TensorFlow neural image captioning training, generate the image captions with a tokenizer and start token, then evaluate the model using Bleu score from nltk.

Assess image captioning quality using Bleu score and generate captions for a sample image, then save tokenizer, max length, and the VGG model for a Streamlit deployment.

Create a Streamlit app to load pre-trained TensorFlow models and tokenizer assets, preprocess inputs with VGG16, and generate image captions using a saved max length and pickle tokenizer.

Build a Streamlit app that uploads images and generates automatic captions using a TensorFlow Keras model with VGG preprocessing and tokenization, including model parameters and caption output.

Launch a Streamlit app with app.py and test it with images using OpenCV and TensorFlow. See captions generated automatically and learn how to deploy the app on AWS EC2.

Launch an Ubuntu-based AWS EC2 instance, configure security for port 8501, and deploy a Streamlit app with Miniconda. Use a higher-end instance for production.