Welcome to the project on How to build low-latency deep-learning-based flask app. In this project, we will refactor the entire codebase of the project How to Deploy an Image Classification Model using Flask. That monolithic code will be refactored to form two microservices - the flask service and model service. The model service acts as a server that renders pretrained Tensorflow model as a deep learning API, and keeps listening for any incoming requests. The flask service requests the model service, and displays the response from the model server. This way, we write cleaner code and promote service isolation.
Further, we will introduce an engineering technique, wherein we introduce the concept of asynchrony using ZMQ networking library, in order to reduce deep learning api response time and thus make the app faster in return. By integrating ZMQ with this client-server architecture, we improve the latency of this microservice-based app. Upon completing this project, we would optimize inference time to classify the input image, and make it a low-latency web application, thus making it deployable in real-time production environments.
Blog : Improving the Performance of Deep-Learning based Flask App with ZMQ
This project depends on another project: How to Deploy an Image Classification Model using Flask
Skills you will develop:
Welcome to this project on Image Classification with Pre-trained InceptionV3 Network. This project aims to impart the knowledge of how to access the pre-trained models(here we get pre-trained Inception model) from Keras of TensorFlow 2, and appreciate its powerful classification capacity by making the model predict the class of an input image.
Understanding the pre-trained models is very important because this forms the basis of transfer learning. one of the most appreciated techniques to perform the classification of a different task thus reducing the training time, the number of iterations, and resource consumption. Learning about the pre-trained models and working hands-on with such models is thus very crucial in deep learning, and the same is the aim of this project.
Skills you will develop:
Welcome to the project on Hosting an Image Classification App on Heroku. In this project, we will get a basic understanding of how to deploy a web app on Heroku, a Platform as a Service.
Heroku is a cloud platform for the deployment and management purposes of web applications. It could be considered as one of the best solutions for hosting web-apps very quickly, thus allowing the developer to concentrate more on development.
In this project, we will learn how to predict images from their noisy version. We will use the MNIST dataset for this project. First, we will load the dataset, explore it, and they we will learn how to introduce noise to an image. Next we will train a KNN Classifier to predict the original image from it's noisy version.
Skills you will develop:
Welcome to this project on Building a CNN Classifier using TensorFlow 2 for MNIST Fashion Dataset. In this project, we will understand how to use the TensorFlow 2 platform to build a simple classifier using Convolutional Neural Networks(CNNs).
CNNs have been one of the state-of-art tools in the current era of Computer Vision. The present-day deep learning and computer vision communities find numerous applications of CNNs in classification tasks and object detection use cases. These use cases have formed the basis of many real-time applications like self-driving cars, face recognition apps, satellite photo analysis and classification, and many more. Thus, it is very important for a Computer Vision aspirant to learn how CNNs work practically, which is the aim of this project.
