googlecreativelab / teachable-machine-v1

Explore how machine learning works, live in the browser. No coding required.

Teachable Machine

About

Teachable Machine is an experiment that makes it easier for anyone to explore machine learning, live in the browser – no coding required. Learn more about the experiment and try it yourself on g.co/teachablemachine.

The experiment is built using the TensorFlow.js library.

We have also released a boilerplate version of this project that can be used as a starting point for your own projects: googlecreativelab/teachable-machine-boilerplate

Development

Install dependencies by running (similar to npm install)

yarn

Build project

yarn build

Start local server by running

yarn run watch

Code Styles

• There’s a pre-commit hook set up that will prevent commits when there are errors
• Run yarn eslint for es6 errors & warnings
• Run yarn stylint for stylus errors & warnings

To run https locally:

https is required to get camera permissions to work when not working with localhost

1. Generate Keys

openssl genrsa -out server.key 2048
openssl req -new -x509 -sha256 -key server.key -out server.cer -days 365 -subj /CN=YOUR_IP

2. Use yarn run watch-https
3. Go to https://YOUR_IP:3000, then accept the insecure privacy notice, and proceed.

Credit

This is not an official Google product, but an experiment that was a collaborative effort by friends from Støj, Use All Five and Creative Lab and PAIR teams at Google.

teachable-machine-v1's Issues

Can this be deployed to a shared host?

I'm a newbie to this and am trying to deploy to a domain on a shared hosting plan.

Is this possible? If so, are there any steps to guide me?

Thanks!

How can I add one or more new classes to train?

It's only 3 classes in the origin, then I want to train more classes.
Is it possible?
I found the green, purple, orange, and red in the code, but the red one wasn't used, maybe we can use it?
But I'm poor in it, so I don't know how to modify this project to train more.
May anyone help me with this?
thx.

Confidence of trained classes not updating when an unknown object appears

Once i trained the model if i put an unknown object in front of the camera the confidence of last detected class not changing from 100 percent. I mean ideally when it detects an unknown object the confidence of the classes should be 0 percent. please let me know if i am wrong here ?

installation problem

Hello,
I want to know how can i install teachable machine v1 in my vps server on my domain name.. do you have any browser based code which can directly be put on my server..? Please guide me for installation ...

how?

How to add artificial intelligence to your site for recognition of movement?

Clicking causes section to restart at "Ok, you should now see the green bar" stage in the tutorial

Steps to reproduce:

1. Complete tutorial up to the step at which the instructor says "Ok, you should now see the green bar (...)"
2. Click anywhere on the screen as that (long) step is being played

At this point, the stage restarts. This is particularly annoying in this longer stage where one may become impatient and begin clicking around.

How to save the training model?

Nice project! But I have a question ...
If I finished three steps, how do I see this trained model after I refreshing the browser?
Did deeplearn provide output model api?

How can i make the gif images localization?

I want to use the local images in my computer as the output rather than the url link and i have already modified the url path which exits in the GIFOutput.js to the local images path ,but i still cannot see anything from the output. How can i do?

Unable to request for camera access when running program

I downloaded the zip and executed the commands to install, build and run. it loaded the page in browser automatically. It's unable to request for camera access. Here the screenshot. I tried clicking on refresh page link but still the same issue

win10 - not working with latest chrome, online or offline . .

Works online with firefox . . .

Gifler not found error

I am trying to build this project
getting error Cannot find module gifler from .....\src\ouputs
Please help in this

Classes show NAN% in pose model

I am facing a nan issue in teachable machine, when I created the pose model and added an input to it, in all the classes the output shows NAN%

Model Link: https://teachablemachine.withgoogle.com/models/IAGVleru7/

This is the output

I tried clearing the cache, also tried running it in incognito mode and also on different browsers. Kindly help me in understanding why this issue occurs and what should be fix for this. Thanks in advance

issue with yarn

I typed yarn in terminal. but this error popes out.
ERROR: There are no scenarios; must have at least one.
Do I have to install some other thins?
PS: I run the code on my Ubuntu 16.04 system locally.
Thanks:-)

How to join 2 image models?

Hi! I was wondering, how can I join 2 trained model data together? e.g my friend wants me to help with training AI to recognize some objects. I know we could take and share photos but is there any other clever way? Maybe joining JSON data?

Prototype Pollution in minimist #48

minimist@^1.2.6:
version "1.2.6"
Minimist <=1.2.5 is vulnerable to Prototype Pollution via file index.js, function setKey() (lines 69-95).

How to open other files in output inplace of gif images.?

In output it's displaying gif images, I want to open any document files in place of gif image how to modify the code.

how can i change the sound output ?

hello ,this is tony from taiwan . I'm a college student . I have some question about your projrct-Teachable Machine . I tried use your source code from github to change the sound output. But it doesn't work . How can i do to change the output?

Image based ML models trained on TeachableMachine V2 show 'NaN' prediction on Chrome and Firefox (Windows OS v 7)

Image based ML models trained on TeachableMachine V2 show 'NaN' prediction on Chrome and Firefox (Windows OS v 7).What could be the problem? The models do not work after training.No errors shown while or after training also.

Duplicated css properties in recording.styl

Duplicated css properties on teachable-machine-v1/style/components/recording.styl

Line 14 opacity
Line 93 Position
LIne 99 width

Cannot find module 'gifler'

Hello,
I am trying to run the code, but I got an error which it can find module gifler.

Could you please let me know how to add gifler module in npm.

Cheers,

Exporting the trained model

Hi guys,

You have said you are releasing a more awesome teachable machine tool: https://teachablemachine.withgoogle.com/io19 and i can't wait to see that.

I have emailed to get access to a BETA version, and i hope i do :) but wanted to know if there is any way of exporting the trained model with the boilerplate at the moment or with this version.

I want to be able to train the system to recognize a specific object, but have no idea how to export the model from the current web tool.

Any help would be greatly appreciated.

Thank you,

Alex

a amazing question

when i first press the button train_green,it is only take one picture.And then I push the button again and nothing happens.the only way to let it record another picture is press the button record a video and close it.
as i spend almost 10 mins to input nearly 30+ picture but sadly to find no output.it doesnot work

Input Shape of the Pose Model

I trained a Pose Model for two classes. I trained it for experimenting on some random labels. I want to run inference on this tfjs model but the model accepts float32[?,14739] as input. Usually, all the image models accept input in these formats:

1. Tensorflow format: [batch_size, height, width, channels]
2. PyTorch format: [batch_size, channels, height, width]

The model doesn't accept input in any of the above format. Am i missing something here?

How did I check the input shape?

1. Download the *.zip file of the model
2. Unzip the file.
3. Open netron.app
4. Upload the model.json file on the netron.app

Link to Pose Model: https://teachablemachine.withgoogle.com/models/irJWnCc6c/

It prompts me to download and use Chrome, while already using Chrome.

I do not have ad blockers installed and WebGL Draft Extensions is enabled via 'chrome://flags/':

Error in Exporting Model (Error:413)

Bug Description

Exporting TensorFlow lite model through EdgeTPU conversion type results in a prolonged process of conversion and results with
displaying

" Something went wrong while converting "

and a broken model

https://drive.google.com/file/d/1O3HhCxMbqixNShY_eVZraVn49MyqEicq/view?usp=sharing

Is there any other way to export my model as EdgeTPU Tflite model

This is the drive link to my project

https://drive.google.com/drive/folders/1ZoDo504F0sL0PYXB972Ronwl4yLh-qFO?usp=sharing

Screenshots

Desktop :

OS: Linux Ubuntu 20.04
Google Chrome
Version 88.0.4324.150 (Official Build) (64-bit)
Additional context

Please rectify the problem and eradicate the bug as soon as possible

Model is not updated even upload to cloud is succeeded

Repro steps

1. https://teachablemachine.withgoogle.com/train/image
2. Create two classes of rock, paper
3. Train model
4. Export model -> Upload model
5. copy url and open it in another browser tab. You can see there's only 2 classes of rock, paper
6. Now in teachable machine, add another class of scissors
7. Train model
8. Export model -> Update cloud model
9. copy url and open it in another browser tab.

Expected

• 3 classes of rock, paper, scissors are shown

Actual

• 
• but the metadata.json has 3 classes
• after few hours (even 24 hrs), it'is not reproduced and shows 3 classes correctly

Any idea how to add serialport to this project?

Hi,
I am not quite familiar with JavaScript, but I and Damon [one of my friends] thought it would be even more amazing if we can add serial port to the output of the system so that we can send some characters to the Arduino.

I tried something on npm, but constantly get an error to import serialport to the output class.

Any idea or thought?

Thanks
Mohsen

Readme link appears to be wrong.

In the project description the project is correctly listed to be at https://teachablemachine.withgoogle.com. In the readme file there is a link to https://experiments.withgoogle.com/ai/teachable-machine which hits a 404.

ai teachable machine

can anyone teach me how to edit more commant such as jump or other comman(instructions).
like put more function to a robot not only left, right , center

Any instrucctions to run with PM2 or similar?

First of all, kudos on this amazing machine learning example.

We are working on localizing it for LATAM, it's an invaluable resource that we want to use it to show what machine learning is capable of doing.

Since we are expecting some traffic, budo doesn't seem like the right choice, so we are trying to serve it using PM2, with no success.

Do you have any suggestions/alternatives?

broken <script src="https://storage.googleapis.com/learnjs-data/deeplearn-latest.js"></script>

https://storage.googleapis.com/learnjs-data/deeplearn-latest.js is access forbidden

Can't run teachable machine, on ubuntu 16.04 - asus GL 552 VW

i wan't to run teachable machine on my laptop (ubuntu GNOME 16.04 - ASUS ROG GL-552VW), but i found this error code :

/home/gnome/Downloads/machine/node_modules/stylus/bin/stylus:715
if (err) throw err;
^

Error: ENOENT: no such file or directory, open 'public/style.css'
npm ERR! code ELIFECYCLE
npm ERR! errno 1
npm ERR! [email protected] style-build: stylus -c -u nib style/main.styl -o public/style.css
npm ERR! Exit status 1
npm ERR!
npm ERR! Failed at the [email protected] style-build script.
npm ERR! This is probably not a problem with npm. There is likely additional logging output above.

npm ERR! A complete log of this run can be found in:
npm ERR! /home/gnome/.npm/_logs/2018-06-30T06_37_18_031Z-debug.lo

what should i do? can someone help he.
-sorry about my english

Error running yarn run watch

App gets stuck after training on ChromeOS

Using ChromeOS, after training the model the app doesn't seem to respond to input and gets "stuck" with the same confidence and output that it had calculated during the last training click. I've tested on multiple Chromebooks and they all have this issue. Using the Chrome browser on Windows (7 and 10) things work as expected.

Will you be open sourcing V2?

I saw that there is a new version of Teachable Machine with Pose Estimation and other goodies. Will you be open sourcing that too? I see that the Community repository has some related code, but not enough to get clueless people started building forks of Teachable Machine

Browser not supported

Hello,

I installed the newest version of google chrome and it is not supported.

I have tried all other browsers such as Brave and Safari but none of them work.

Kindly assist me.

QUESTION: How secure is it to train with sensitive images? Is it all done locally?

I didn't see any uploading in the network tabs and have not found anything yet in the code where the images were being uploaded anywhere. Is the training all done locally?

I was wanting to train the model to detect blurry images of documents, but the training material has personally identifying information and I want to me 100% sure its safe

Camera access blocked on Windows10 chrome

I'm facing this weird issue on Windows 10 chrome, chrome doesn't allow camera to be accessed and this project mainly needs camera to work. I get Uncaught Error: This browser does not support WebGL. in the console.

This worked fine on Mac but not on windows 10.

A saved model now only picks one category when it worked well about a month ago

This model distinguishes between my left and right hand: https://teachablemachine.withgoogle.com/models/X3b80T2Ho/

It now reports either hand as right and yet it worked a few weeks ago. Both were trained with the same lighting, background etc just a minute apart. Flipping the camera input image has no effect.

I just retrained a model in the same way and this one works fine: https://teachablemachine.withgoogle.com/models/0O5MUcOjY/

Why would the old model stop working?

I can't open this web page in China

https://teachablemachine.withgoogle.com/

I can't open this web page in China

I am unable to export

I have been trying to export a quantized model, each time I get a something went wrong error.
This is very urgent and time sensitive, please advice.

I have been trying to use node net package in your output GIFOuput.js .It gives error net.connect is not found