Skip to content

Users' Guide

Cody Neiman edited this page Feb 23, 2018 · 5 revisions

Note: This application requires an active internet connection

Usage

When opening the application you will be brought to this start screen:

Open Screen

The settings button when clicked slide-toggles the settings options which must all be filled before you can progress further:

Settings

The log button when clicked slide-toggles a log of recent debugging information:

Log

This field is not necessary if you are wanting to load a previously created neural network.

First, we must click the browse button next to the label "Photos Directory" in the settings to browse to the folder holding your training images. (You may also manually enter the directory in the input field)

At this time, images must be .jpg (.png support planned in a future release)

The names of the directories and files must have no spaces in them and directory must be in this format: (file names do not need to be in any specific format)

--training_images_folder
----category1
------imageOfCategory1_001.jpg
------imageOfCategory1_002.jpg
------imageOfCategory1_...
----category2
------imageOfCategory2_001.jpg
------imageOfCategory2_002.jpg
------imageOfCategory2_...
----category...

Train Tree

Next, we must select where to save the neural network's files and bottlenecks using the browse button next to the label "TF Files Directory" (standing for TensorFlow Files Directory). (You may also manually enter the directory in the input field)

If you are wanting to create a new neural network, select an empty folder; if you are wanting to load a previously created neural network select the folder you selected last time in this step that holds your TensorFlow files.

Again, the names of the directories must have no spaces in them.

This is what a folder looks like after a neural network is created in it:

Neural Network Files Tree

If the previous steps are successful you should now see a new option appear: "Start TensorBoard"

Start TensorBoard

Click on start TensorBoard to begin monitoring for the neural network. You can read more about TensorBoard here.

You are now able to view TensorBoard by going to: localhost:6006

*This feature currently only works on Linux (Windows support planned in a future release)

When you click this button you might notice this image:

Loading

This .gif signifies that the application is attempting to load something. There is currently no hang detection so if you are suspicious of the status, check the log for more information.

If you where attempting to load a neural network you will be now greeted with the test picture option.

Once TensorBoard has been completed the loading animation will go away and you will be greeted with additional options:

TensorBoard Running

The first option "Stop TensorBoard" will attempt to close the TensorBoard process but it may fail depending on your system. (better functionality planned in a future release)

If it succeeds, it will return you to the previous page from before starting TensorBoard.

The second option "Create Neural Network" will attempt to create a neural network based on the supplied parameters.

Once pressed, a loading bar will appear showing you the step it is on and the progress through that step.

The first step is "Downloading required assets..." where the program will download the needed MobileNet version and other required assets.

Downloading Required Assets

Once that step has completed the second step "Creating bottlenecks..." will begin. You will again see a loading bar reflecting the status of the step. This step will likely take the longest time to complete.

Creating Bottlenecks

Once that step has completed the third and final step "Training neural network..." will begin. Again, you will see a loading bar reflecting the status of the step. This step will likely take the 2nd longest amount of time to complete.

Training Neural Network

After all three steps have completed you will be given an additional option "Test Picture" and the loading status will be removed.

Test Picture

This button when pressed will allow you to browse to an image to test through the neural network. (Remember, only .jpg supported)

Once you select an image it will be ran through your neural network and the results will be displayed.

Results

You have successfully used ez-nn! Your neural network files have been saved to the "TF Files Directory" you specified and can be loaded any time you like by opening this application, selecting the "TF Files Directory" and the option to test a picture should appear after starting TensorBoard.

More Information

At various times during the process the log will show relevant information that may also be useful in debugging if something goes wrong.

When creating bottlenecks:

Log During Creating Bottlenecks

When training the neural network:

Log During Training of the Neural Network

Cleanup

Before closing ez-nn to prevent ram leakage, you should close TensorBoard (unless of course, you want it to remain open).

If the button to close TensorBoard does nothing when pressed, close ez-nn and search for "tensorboard" in your process explorer of choice and kill the process.

You might be able to find out more information in the log as to why it was unsuccessful in closing TensorBoard.

If things go wrong, you should first open the log and see if you can decipher the errors.

Troubleshooting: Common Problems and Solutions

Permission Related:

During Installation:

  • You may need to switch to root to install by npm and yarn. This is not typical for installations and is currently being investigated.

During Usage:

  • You might have selected folders or files that are already currently in use. Restarting the computer or selecting different folders may fix this problem.

Invalid Directory:

During Usage:

  • Check to make sure your directories do not include any spaces or odd symbols. If you are using letters, dashes, and underscores this is not the problem.

Other:

  • If the solutions above do not work or you are facing a different problem, file an issue report here