Are you someone who's getting interested in computer vision or any state-of-the-art knowledge in deep learning? Did you know that Tensorflow is an open-source end-to-end platform that is being developed by the Google Brain team which was led by the Google senior fellow and AI researcher Jeff Dean built in November 2015. It can actually perform various tasks focused on training and inference of deep neural networks. This allows the developers to create better machine learning applications using the tools, libraries and community resources. In fact, it is one of the most known deep learning libraries globally which is Google's Tensorflow.

Before you read this article you should also read my first article here, Because in my first article I have given you the overview of my project and the motivation to do this project, If you have already read part 1 it's time to get started. In this article part2, I will focus on the steps and processes to build an image classification model. This dataset has 3017 images of 16 different mushrooms, and they also classify those 16 mushrooms as edible and Not edible. Since I use google collab to train this model, I need to download the data set in my google drive. I use google collab because I can train my model on GPU that was provided by Google.

If it is the first time you create a project, you will be directed to create a new project. This will take about 1–2 minutes. And paste it inside the tab where the JupyterLab is opening. This has all the necessary Tensorflow libraries for building a custom AI model installed. You will see a new "dataset" folder created as well as a new cats-and-dogs.zip"

Generating training and validation batches with tf.data is way faster than ImageDataGenerator. Let's compare; First, we use ImageDataGenerator without using any augmentation -- Now we will use tf.data -- So tf.data is about 34 times faster than ImageDataGenerator and one of the main reasons for that is a technique called'Prefetching'. In the TensorFlow documentation series of examples are given with excellent explanations but here in brief what is happening. The data pipeline can be thought of as a combination of'producer' (generating batches) and'consumer' (batches that are used to train the neural net) and prefetch transformation provides benefits whenever there is an opportunity to overlap the work of a'producer' with the work of a'consumer'.

Transfer learning is an important topic. As a civilization, we have been passing on the knowledge from one generation to the other, enabling the technological advancement that we enjoy today. It's the edifice that supports most of the state-of-the-art models that are blowing steam, empowering many services that we take for granted. Transfer learning is about having a good starting point for the downstream task we're interested in solving. In this article, we're going to discuss how to piggyback on transfer learning to get a warm start to solve an image classification task. The content of this article is based on "TensorFlow 2 in Action" by Manning and on TensorFlow 2.2.

CNNs have broken the mold and ascended the throne to become the state-of-the-art computer vision technique. Among the different types of neural networks (others include recurrent neural networks (RNN), long short term memory (LSTM), artificial neural networks (ANN), etc.), CNNs are easily the most popular. These convolutional neural network models are ubiquitous in the image data space. They work phenomenally well on computer vision tasks like image classification, object detection, image recognition, etc. So – where can you practice your CNN skills?

ImageDataGenerator is Keras's go-to class for pipelining image data for deep learning. It allows easy access to your local file-system and multiple different methods for loading in data from different structures. It also has some pretty powerful data pre-processing and augmentation capabilities. For the purposes of this tutorial, we won't be doing much data augmentation, we will primarily be focusing on the different methods for reading data in using ImageDataGenerator. If you already have your own image data simply need a quick tutorial on a single method, review the'Methods and use-cases' section, and continue down to the appropriate tutorial.

Computer vision has many uses. It can recognise faces, it can be used in quality control and security and it can also recognise very successfully different object on the image. Today we will look at the last example. We will build a supervised machine learning model to recognise cats and dogs on the image using Neural Networks. You will learn how to create and configure a Convolutional Neural Network (CNN).

Now we calculate the threshold value for every frame and determine the contours using cv2.findContours and return the max contours (the most outermost contours for the object) using the function segment. Using the contours we are able to determine if there is any foreground object being detected in the ROI, in other words, if there is a hand in the ROI. When contours are detected (or hand is present in the ROI), We start to save the image of the ROI in the train and test set respectively for the letter or number we are detecting it for. In the above example, the dataset for 1 is being created and the thresholded image of the ROI is being shown in the next window and this frame of ROI is being saved in ..train/1/example.jpg For the train dataset, we save 701 images for each number to be detected, and for the test dataset, we do the same and create 40 images for each number. Now on the created data set we train a CNN. First, we load the data using ImageDataGenerator of keras through which we can use the flow_from_directory function to load the train and test set data, and each of the names of the number folders will be the class names for the imgs loaded.

This line of code is used to define the transformations that the training DataGenerator will apply on all the images to augment the size of the dataset. For the validation DataGenerator, we only specify the scaling factor. The other transformations are not required because we are not training the model on this data. Next, we define the Model. We set layer.trainable False for each layer of the VGG model, as we are using the pre-trained weights of the model.