Welcome to the project on Training from Scratch vs Transfer Learning. In this exercise, we will understand how to train a neural network from scratch to classify data using TensorFlow 2. We would also learn how to use the weights of an already trained model to achieve classification to another set of data.
We will train a neural network (say model A) on data related to 6 of the classes, and we will train another neural network (say model B) on the remaining 2 classes. Then, we would use the pre-trained weights of model A and tune the last layer so as to classify these 2 classes(this technique is called Transfer Learning), and compare the results of classification obtained using normal training and transfer learning. In this project, we would practically appreciate the use of Transfer Learning.
Welcome to the project on Working with Custom Loss Function. This project aims to provide an understanding of how we could use the custom defined loss functions along with TensorFlow 2.
Though TensorFlow 2 already provides us with a variety of loss functions, knowing how to use a user-defined loss function would be crucial for a machine learning aspirant because often times in real-world industries, it is expected to experiment with various custom defined functions. This exercise is designed to achieve that goal.
Skills you will develop:
Welcome to this project on the Neural Style Transfer. In this project, you will use TensorFlow 2 to generate an image that is an artistic blend of a content image and style image.
Neural Artistic Style Transfer finds a wide range of applications to fancily modify images. This field has so much influenced the technical world that many apps, such as Prisma, have received great craze amongst the users. In recent days, decent work has also been done in this area, which served as a holy grail to our project. The heart of this capability is the convolutional neural network, often used for modern image processing. The work of Leon A. Gatys, et al on ”A Neural Algorithm of Artistic Style”[1], has first come up with such an approach using a powerful pre-trained VGG19 model trained on the famous ImageNet database images.
Skills you will develop:
Welcome to this project on Credit Card Fraud Detection. In this project, you will use Python, SMOTE Technique(to over-sample data), build a Logistic Regression Classifier, and apply it to detect if a transaction is fraudulent or not.
The real world datasets often might be with data of imbalanced classes. It is very important to feed a decent number of data samples of each class in a classification problem so that the classifier would detect the underlying hidden patterns for each class and prepare itself to reasonably classify the test data. Upon completing this project, you will understand the pragmatic application of various Pandas functions, with a clear picture of how to over-sample the dataset with imbalanced classes using the SMOTE technique and how to use the thus obtained data to train a classifier.
Skills you will develop:
Welcome to this project on Image Stitching using OpenCV. In this project, we will use OpenCV with Python and Matplotlib in order to merge two images and form a panorama.
As you know, the Google photos app has stunning automatic features like video making, panorama stitching, collage making, and many more. In this exercise, we will understand how to make a panorama stitching using OpenCV with Python.
Skills you will develop:
