Azure Machine Learning Service - deploy to webservice

Tensorflow Object Detection API - Containerize the web service

Tensorflow Object Detection API - Inference

After you get a machine learning model by training on Tensorflow, you can run that model and receive the detection result. The source code is here on Github. (https://github.com/koheikawata/objectdetectiontest). Below are examples of inputs and outputs of "object_detection_inference_test1.py"

Precision and Recall

Precision




Recall

Tensorflow Object Detection API - TFRecord creation

Understand how to create TFRecords is an important point when you use Tensorflow Object Detection API. In my Github repo (https://github.com/koheikawata/objectdetectiontest), you can find what directory structure looks like.

``` . └── research ├── object_detection | ├── legacy | | ├── train.py | | └── eval.py | └── test1 | ├── eval_images | | └── export-image-0-b'IMG_0000'.png | ├── checkpoints | | └── model.ckpt-4495.data-00000-of-00001 | | └── model.ckpt-4495.index | | └── model.ckpt-4495.meta | ├── output | | └── saved_model | | └── frozen_inference_graph.pb | ├── Annotations | | └── ***.xml | ├── JPEGImages | | └── ***.jpg | ├── flask_pics | ├── pascal_label_map.pbtxt | ├── train.record | └── val.record ├── faster_rcnn_resnet101_coco_2018_01_28 | └── model.ckpt ├── faster_rcnn_resnet101_coco_test1.config ├── create_tf_record_test1.py ├── object_detection_app_test1.py ├── object_detection_inference_test1.py ├── object_detection_app_test1_docker.py └── Dockerfile ```

Classification --- Deep Learning

We can try a deep learning algorithm by utilizing Keras on Tensorflow or CNTK. However, understanding what is going on inside neural networks is not an easy task. Visualizing how input data is converted into a simple array that describes classification category helps you figure out its structure.

Keras deep learning training input

Keras allows you to use neural networks easily. It is very easy to try a deep learning prototype with Tensorflow or CNTK. The training starts with just one command like below.

``` model.fit(x_train, y_train, epochs=5, batch_size=32) ```

Understanding what is "x_train" and "y_train" is a starting point.

x_train

x_train is multi-dimensional array including numerical elements of target contents. For example, when you train 1000 images with 85 * 128 pixels RGB, x_train should be a 4-dimensional array (1000, 85, 128, 3).
x_train datatype is normalized between 0 and 1 with float32 in this case. (RGB is described as 8 bit)

``` x_train = x_train.astype('float32') x_train /= 255 ```

Good articles for machine learning

RBF SVM (Radial Basis Function Kernel Support Vector Machine)
https://www.youtube.com/watch?v=3liCbRZPrZA&feature=youtu.be

Classification: Precision and Recall
https://developers.google.com/machine-learning/crash-course/classification/precision-and-recall

Classification --- Deep Learning
https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/

Python setting up errors

I am using Azure DSVM Standard DS12 v2 (4 vcpus, 28 GB memory) with Linux, which pre-installed Anaconda and related machine learning tools. It is sometimes confusing as Python 2 and 3 are being used through Jupyter notebook in one machine, and I would like to keep memos when setting up errors happen.

Image Equalization and Normalization

Preprocessing data before training is a very important and critical part for machine Learning. In computer vision, Image Equalization and Normalization are confusing, but very different. I want to make clear out the difference.

In this article, I am going to normalize and equalize the pic below that Samurai visited Egypt in 1864. This image is really old and blurry. When you normalize and equalize, the image becomes clearer to see.

Azure Batch AI

Today I tried Azure Batch AI and want to share something useful. The most complex part for me was pricing. The official documentation mentions there is no additional pricing by Azure Batch AI, but it is important when and how it charges us.