Explore
For Enterprise
Join for Free
Log In

閲覧
検索
企業用
ログイン
参加は無料

Working through the notebook

Loading...

Convolutional Neural Networks in TensorFlow

deeplearning.ai

4.7 (1,966 件の評価) | 29K在籍する学生

TensorFlow in Practice 専門講座における 4 の 2 コース

If you are a software developer who wants to build scalable AI-powered algorithms, you need to understand how to use the tools to build them. This course is part of the upcoming Machine Learning in Tensorflow Specialization and will teach you best practices for using TensorFlow, a popular open-source framework for machine learning. In Course 2 of the deeplearning.ai TensorFlow Specialization, you will learn advanced techniques to improve the computer vision model you built in Course 1. You will explore how to work with real-world images in different shapes and sizes, visualize the journey of an image through convolutions to understand how a computer “sees” information, plot loss and accuracy, and explore strategies to prevent overfitting, including augmentation and dropout. Finally, Course 2 will introduce you to transfer learning and how learned features can be extracted from models. The Machine Learning course and Deep Learning Specialization from Andrew Ng teach the most important and foundational principles of Machine Learning and Deep Learning. This new deeplearning.ai TensorFlow Specialization teaches you how to use TensorFlow to implement those principles so that you can start building and applying scalable models to real-world problems. To develop a deeper understanding of how neural networks work, we recommend that you take the Deep Learning Specialization.

シラバスを表示

学習するスキル

Inductive Transfer, Augmentation, Dropouts, Machine Learning, Tensorflow

レビュー

4.7 (1,966 件の評価)

5 stars
1,542 ratings
4 stars
316 ratings
3 stars
86 ratings
2 stars
11 ratings
1 star
11 ratings

MH

May 24, 2019

A very comprehensive and easy to learn course on Tensor Flow. I am really impressed by the Instructor ability to teach difficult concept with ease. I will look forward another course of this series.

CM

May 01, 2019

A patient and coherent introduction. At the end, you have good working code you can use elsewhere. Remarkably, the primary lecturer, Laurence Moroney, responds fairly quickly to posts in the forum.

レッスンから

Exploring a Larger Dataset

In the first course in this specialization, you had an introduction to TensorFlow, and how, with its high level APIs you could do basic image classification, an you learned a little bit about Convolutional Neural Networks (ConvNets). In this course you'll go deeper into using ConvNets will real-world data, and learn about techniques that you can use to improve your ConvNet performance, particularly when doing image classification! In Week 1, this week, you'll get started by looking at a much larger dataset than you've been using thus far: The Cats and Dogs dataset which had been a Kaggle Challenge in image classification!

A conversation with Andrew Ng1:20

Training with the cats vs. dogs dataset2:51

Working through the notebook4:50

Fixing through cropping0:49

Visualizing the effect of the convolutions1:40

Looking at accuracy and loss1:26

Week 1 Wrap up0:33

講師

Laurence Moroney
AI Advocate

字幕

The first thing we will do is download the set of images. They're stored as a zip file containing 3,000 images, 2,000 of which we'll use for training, and 1,000 for testing. Once they're downloaded, you'll need access to the underlying operating system of the virtual machine on which this Colab runs. This is available in the OS namespace. This code will unzip the cats and dogs data that you just downloaded, into the /tmp directory. In there, the sub-directories will be recreated, because they're stored that were in the zip file. The next thing is for us to set up our directories as variables, so we can point the generators at them. Now, that we have the directories as variables, we can pass them to the os.list here to take the files from those directories, and load them into Python lists. We can see there's a list of file names like cat253.jpg, etc. On the files tab, we can also inspect the file system and in TMP, we can see our cats versus dogs folder, within which we have training and validation sub-directories, which each contain cats and dogs directories. The images, of course, are stored in there. If we then count the images in each directory, we can be sure that we have the right amounts of images. As we can see, we have a thousand of each animal in Training and 500 of each in Tests for total of 3000. Next up, we can visualize some of the data, so we can see how diverse it is. This code would set a matplotlib which is a Python library for drawing graphics. This code will pick up some random cats and dogs and draw them in a grid. We can then see once it's drawn that there's a lot of diversity in these images. There's different colors of animal, there's different location within the picture, and there's even sometimes multiple items in the picture, like the lady here holding a cat. We can then run it again to see some more. This image containing multiple cats is particularly challenging. Okay. Now, let's build our neural network. We'll import TensorFlow, and then we'll define our model. We'll print the summary, and here you can see the output shape of how the image passed through the layers, and gradually reducing in size through the convolution and pooling. Here, is where we compile our model, defining the loss function and the optimizer. Here, is where we set up the two generators, pointing them at the training and validation sub-directories. These contains sub-directories of their own, each with cats and dogs. When I run it, you'll see the printout. It found 2,000 images in two classes, and that's the training, and 1,000 images in two classes, that's the testing. Now, we can do the training. This time instead of using model.fit, we'll again use model.fit generator because our data source is the generators. Now, you can see they're passed in. Now, we'll start the training and we'll watch it progress. It should take 3-3.5 minutes. When it's done, you can see the accuracy is about 73% and it's not bad. It's not great but it's not bad. Okay. So let's take a look at how the model predicts on some images. So I'm going to choose these five images from my hard drive and upload them. Once they're uploaded, you can see that the classifier gives me a prediction on each. So let's now compare. We can see that the first one is obviously wrong but let's now look at each image to see if we can compare the results. This first one is really impressive and as you can see, the dog is in a very small part of the image and there's lots of other features like trees, mountains, skies and lakes, and the dog also has its face turned away, but the classifier still got it right. This one is obviously a dog, but again it's partially obscured. You can't see the body. It's just a white mass and the face while pronounced, is also hidden amongst the fur, but it got it right. This one's a bit more obvious, but it's also a very different looking dog than the previous one. The eyes, nose and mouth are all part of a dark patch, but the neural network still recognize the dog. This one's also very impressive. There are two cats in the image and one is mostly hidden, but that didn't confuse the model that correctly recognized the cat here. Strangely enough though, this is the one that the model got wrong. Maybe it's because we can only see one eye, or maybe the sharply defined branches to the left confused the model. So take a look, think about what you would do to go through this workbooks at this point and have some fun with it. Find some images of cats and dogs and upload them to the notebook and get it to classify them, then see if you can find one that you would expect to be easy that the prediction gets wrong. Once you've done that, see if you can edit that image in a way that gets it to work.

コース一覧で検討

サインアップは無料です。今すぐサインアップして、パーソナライズされたお勧め、更新、サービスを利用しましょう。

Coursera

このコースについて
リーダーシップ
キャリア
カタログ
修了証
MasterTrack™の修了証
学位
企業用
政府向け
キャンパス向け

コミュニティ

受講生
パートナー
開発者
ベータテスター
翻訳者
ブログ
技術ブログ

さらに表示

規約
プライバシー
ヘルプ
アクセシビリティ
報道関係者
お問い合わせ
ディレクトリ
アフィリエイト

場所を選ばす学習する

フォローする

©＃{currentYear} Coursera Inc.All rights reserved.