Advanced-Deep-Learning-and-Computer-Vision

This is a course from AI Engineer program from SimpliLearn. In this course, we will be able to classify different advanced models of machine learning:

Boltzman machines/RBM/DBNs
Various types of Auto-Encoders
Different types of GAN models

Also, we will practvie on certain applications such as:

NST - Neural Style Transfer
StyleGANs
Various type o fother GANs
YOLO/OpenCV based project
AutoEncoders/GANs for data generation

Additional focus: Deep Learning Model Deployment, Distributed computing using Tensorflow (V2) / Keras, Introduction to Reinforcement Learning

Fundamental Definitions/Terms:

• Advanced Deep Learning: It is a field of study that deals with the recent advancements in deep learning.
• Need of Advanced Deep Learning: To stay up-to-date with the recent advancements happening in deep learning, there should be a dedicated field of study.
• Computer Vision: Computer vision is an interdisciplinary scientific field that deals with how computers can gain high-level understanding from digital images or videos.
• Applications of Computer vision:

  Traffic Monitoring, Interpretation of High Resolution Computer Images
  Optical Character Recognition, Target Recognition, 3D Shape Reconstruction
  Face Detection
  

Practice Exercises:

• Image Pre-processing (Data Manipulation) - Week Folder, Notebook
• RBM/Autoencoders- Week Folder, Notebook - RBM, Notebook - Autoencoders Part 1, Notebook - Autoencoders Part 2, Dataset
• Generative Adversarial Networks - I (GANs) - Week Folder, Notebook - GAN Part 1
• Generative Adversarial Networks - II (GANs) and Neural transfer / Object Detection - Week Folder, [Notebook 1 DCGAN CIFAR10], [Generated Image], [Output .h5 File], [Notebook 2 CGAN FASHION MNIST], [Output .h5 File]

Assisted Practice Projects (To be completed)

• Build an Movie Recommendation System Using RBM

• Use Variational Autoencoder with Tensorflow to generater images using MNIST dataset

• Use Variational Autoencoder with Keras to generate images using the MNIST dataset

• Use Keras or TensorFlow to build a deep generative model that will translate drawings of shoes to designs.

• Use YOLO v3 pretrained model for object detection

Final Project

Perform Facial Recognition with Deep Learning in Keras Using CNN

Description

Problem Statement: Facial recognition is a biometric alternative that measures unique characteristics of a human face. Applications available today include flight check in, tagging friends and family members in photos, and “tailored” advertising. You are a computer vision engineer who needs to develop a face recognition programme with deep convolutional neural networks.

Objective: Use a deep convolutional neural network to perform facial recognition using Keras.

Dataset Details: ORL face database composed of 400 images of size 112 x 92. There are 40 people, 10 images per person. The images were taken at different times, lighting and facial expressions. The faces are in an upright position in frontal view, with a slight left-right rotation.

Details

• Program Code + Dataset

  • In the program, we analyzed ORL Faces Dataset where we already had train and test datasets. The image were regenarated using CNN model and accuracy is tested against test data.
  • The analysis for different activation functions is fisrt observed to find that 'leaky-relu' activation function is one of the activation functions that can be used for out final model.
  • The model training is done using x_train and y_train with validation data as x_valid and y_valid. owever for evaluating model, we use x_test and y_test which gives us loss ~0.2435 with an accuracy of 93.75%.

Results (Figures)