This example shows how to run an image classification model on a given dataset.

The model and the data used originate from TensorFlow Lite Micro MNIST LSTM example.

More information about MicroAI Foundation Library can be found in the ApplicationDeveloperGuide.

• MICROEJ SDK 6.
• A VEE Port that contains:
  • EDC-1.3 or higher.
  • BON-1.4 or higher.
  • MICROAI-2.0 or higher.

This example has been tested on:

• IntelliJ IDEA with MicroEJ plugin for IntelliJ IDEA 1.2.0.
• sim-vee-port VEE Port project (Simulator only).

Follow MICROEJ SDK 6 Installation Guide to set up the SDK.

The sample runs through ten 32x32 bitmap images of handwritten digits, taken from the MNIST database. Each image pixel bitmap will be given as an input to the model which will return the digit prediction.

For simplicity, the original PNG files have been converted to BMP.

Run the following command in your IDE (or click the Play button next to the line below when opening this README in IntelliJ IDEA):

./gradlew :microai.mnist_lstm:runOnSimulator

Alternative ways to run in simulation are described in the Run on Simulator documentation.

The results of the interpreter are logged into the console.

The ASCII representation of the image is for cosmetics only, it is not the actual data given to the model interpreter.

----------------------------
----------------------------
----------------------------
----------------------------
----------------------------
----------------------------
----------------------------
------******----------------
------****************------
------****************------
-----------***********------
------------------****------
-----------------****-------
-----------------****-------
----------------****--------
----------------****--------
---------------****---------
---------------***----------
--------------****----------
-------------****-----------
------------*****-----------
------------****------------
-----------*****------------
-----------****-------------
----------*****-------------
----------*****-------------
----------****--------------
----------------------------
[microai mnist lstm sample] INFO: image: 7 -> prediction: 7 with confidence: 99%

The dependencies defined in build.gradle.kts are configured in libs.versions.toml.

All dependencies are retrieved transitively by Gradle.

N/A

None.

Copyright 2025 MicroEJ Corp. All rights reserved.
Use of this source code is governed by a BSD-style license that can be found with this software.