node-red-contrib-teachable-machine

A Node-RED node based in tensorflow.js that enables to run custom image classification trained models using Teachable Machine tool. All notable changes to this project will be documented in the CHANGELOG.md file.

Install

You have two options to install the node.

• Use Manage palette option in Node-RED Menu (recommended)

• Run the following command in your Node-RED user directory - typically ~/.node-red

  npm install node-red-contrib-teachable-machine

Note: If you run the command you will need to restart Node-RED after installation. If installation goes wrong please open a new issue.

Node usage

Step 1

Go to Teachable Machine and follow the steps to train your custom classification model. Once trained click on the Export Model button.

Step 2

Select Tensorflow.js format and upload your trained model (for free). Once it is uploaded, copy the generated URL.

Once the URL is generated it will show the stored files.

Step 3

Online Mode

Select Online Mode and paste the saved URL in the node configuration. That URL hosts all the information to load your trained model. Make sure you copy all the given URL including the https://... and the / in the end.

Local Mode

Download all three files from the generated URL and save them locally in a folder maintaning the original filenames.

Select Local Mode and write down the absolute path of the folder that contain the three downloaded files in the node configuration. Make sure it ends with a / noting it is a folder.

Step 4

In Node-RED send a buffered image (jpeg or png) to the node. Check the example in the Import section.

Node Status Information

Shape

• ■ dot: node is idle
• □ ring: node is working

Color

• 🟩 green: model is available
• 🟨 yellow: preparing model
• 🟥 red: node error

Requirements

• Node-RED v2.1.0+
• Node.js v16.20.0+

Warning: Only the official Docker nodered/node-red image based on Debian v3.1.0+ image works since it needs ld-linux-x86-64.so.2. This necessary library is not present in the default Alpine Docker image version.

Changelog

All notable changes to this project will be documented in this file following a Semantic Versioning. The format is based on Keep a Changelog.

1.4.0 - 2022-11-30

Added

• Local Mode Support

Changed

• Improved code structure
• Upgrade pre-commit dependencies
• Dependency change from @tensorflow/tfjs to @tensorflow/tfjs-node
• Upgraded node-fetch to version 3.3.0
• Upgraded pureimage to version 0.3.14

1.3.1 - 2021-10-09

Added

• Using pre-commit.ci with its own badge
• New Code Climate mantainability score badge

Changed

• Improved status node information
  • shape:
    • ■ dot: node is idle
    • □ ring: node is working
  • color:
    • 🟩 green: model is available
    • 🟨 yellow: preparing model
    • 🟥 red: node error

Fixed

• Node hungs when using a bad image buffer, even with PNG images - #21

1.3.0 - 2021-09-24

Added

• Model reload option flag during running time - #22 - Thanks @acejacek for the suggestion and the PR
• pre-commit common checks with standard, yamlfmt and mdformat style checks
• GitHub Actions CI test with badge
• npm quality badge to the README file
• Automatic dependency checks with dependabot

Changed

• Upgraded node-fetch to version 3.0.0
• Upgraded pureimage to version 0.3.5
• Upgraded package-lock.json file format to v7
• Updated the basic example with the new reload option flag

1.2.2 - 2020-10-24

Changed

• Upgraded @tensorflow/tfjs-node to version 1.4.0 to help Raspberry Pi installation - #18

1.2.1 - 2020-10-24

Added

• package-lock.json to ensure always exact versions during installation

Fixed

• Output API was changed by mistake, now msg.payload still outputs class and score keys for each classification as in v1.1.0+ releases
• When selecting all predictions without filter only one result was showing up

1.2.0 - 2020-10-23

Added

• Compatibility with official Node-RED Dockerized image based on Alpine image
• Compatibility with Raspberry Pi
• pureimage npm package dependency to manage buffer images using pure javascript - #17
• node-fetch npm package dependency to manage http request to obtain the model info

Removed

• @teachablemachine/image npm package dependency
• jsdom npm package dependency
• canvas npm package dependency

1.1.5 - 2020-09-08

Fixed

• Error when installing nodes - #15

Changed

• Upgrade jsdom version from 16.2.2 to 16.4.0

1.1.4 - 2020-06-05

Fixed

• Better usage of HTMLVideoElement class imported from dom.window
• Better accuracy representation in the node status without decimals

Changed

• Some variable types from const to var

1.1.3 - 2020-06-05

Fixed

• Prediction does not work when save_image's box is checked - #14

Changed

• Dependancy is now @tensorflow/tfjs 1.3.1 instead of @tensorflow/tfjs-node 1.4.0, to match teachable machine correct dependencies

1.1.2 - 2020-05-03

Added

• PayPal donation badge in README file

Fixed

• className key (when Best prediction option) fixed to class in the results
• Refinements on README file
• Alignment HTML icons in node configuration

1.1.1 - 2020-04-24

Added

• New configuration checkbox to pass through the input image in msg.image

Changed

• probability key changed to score in the results
• Updated example with new configuration parameter

Fixed

• Selecting Best prediction option made the Name disappear

1.1.0 - 2020-04-22

Changed

• Updated image on how to use Teachable Machine and configuration node on Step 3
• Use standard image treatment for README instead of HTML
• Upgraded @tensorflow/tfjs-node to version 1.4.0 to enable coexistantce with tfjs-nodes nodes - #8

1.0.1 - 2020-04-15

Changed

• Updated information help node

1.0.0 - 2020-04-15

Added

• Total npm downloads badge
• Online/Local options in configuration node (Local still not functional)
• Internal common functions to set node status
• New Filters configurations when All predictions Output mode is selected
  • Threshold in % - (0 -> 100%)
  • Max. results - (1 -> 5)
• General code optimitzations
• Mentions section in README file

Changed

• Icon updated to Tensorflow 2.0 new logo
• Updated configuration node
• Using all README badges from Shields.io
• Outputs is always an array of results even if Best prediction is selected

0.1.3 - 2020-04-12

Added

• Use of badges in README file
• New images for Installation and Node usage in README
• JavaScript code has been standarized following Standard JS

Changed

• Information on HTML node #5

Fixed

• Typos on README
• basic example with issues to load
• Errors not shown on console or in Node-RED

0.1.2 - 2020-04-12

Added

• Improvements on README file

Fixed

• Loading model error response.arrayBuffer is not a function #3

0.1.1 - 2020-04-11

Added

• Comments in the code

Changed

• Downgraded @tensorflow/tfjs-node from version 1.4.0 to 1.3.1 for better compatibility
• Output has changed from checkbox to a list, you can now select Best predictions or All predictions
• Code cleaning

Fixed

• WebGL loading error in JSDOM
• When installing the node -> npm WARN @teachablemachine/image@0.8.4 requires a peer of @tensorflow/tfjs@1.3.1 but none is installed

0.1.0 - 2020-04-11

Added

• Functional using Teachable Machine Online Model URL
• Option to select Top-1 or all results
• CHANGELOG.md file
• README.md file