Run an Arm AI Portal depth estimation model on Android
Connect an Android phone
Enable Developer options and USB debugging on the Arm-based Android phone. Connect the phone with a data-capable USB cable, unlock it, and accept the debugging authorization prompt.
Verify the connection and architecture:
adb devices -l
adb shell getprop ro.product.cpu.abi
The first command lists the phone as device. The second command returns arm64-v8a.
If the phone is unauthorized, unlock it and accept the prompt. On Windows, you might also need the manufacturer’s USB driver.
Clone the application repository
Clone Image Analysis, which contains the validated Depth Anything V2 adapter, then enter the project directory:
git clone https://github.com/arm-education/ai-portal-android-app-image-to-image.git
cd ai-portal-android-app-image-to-image
Run the remaining terminal commands from this project directory.
The repository contains Kotlin adapters and placeholder files, but it doesn’t contain model weights or credentials.
Prepare the Arm AI Portal model downloader
Create a Python virtual environment and install the package that’s used to download models from Hugging Face:
python3 -m venv .hf-venv
.hf-venv/bin/python -m pip install --upgrade pip huggingface_hub
py -m venv .hf-venv
.\.hf-venv\Scripts\python.exe -m pip install --upgrade pip huggingface_hub
The commands invoke the virtual environment’s Python executable directly, so you don’t need to activate the environment or change the PowerShell script-execution policy.
If Python reports that venv is unavailable on Debian or Ubuntu, install the python3-venv package and rerun the command.
Build and install Image Analysis
Before building, confirm the Java Virtual Machine (JVM) configuration for Gradle:
chmod +x gradlew
./gradlew --version
.\gradlew.bat --version
The Launcher JVM entry should report JDK 17 or later. The Daemon JVM entry should report that Java 17 is configured by gradle/gradle-daemon-jvm.properties. Gradle downloads a compatible JDK 17 automatically when one isn’t already available.
Build the debug APK, then run the unit tests and lint checks:
./gradlew :app:testDebugUnitTest :app:assembleDebug :app:lintDebug
.\gradlew.bat :app:testDebugUnitTest :app:assembleDebug :app:lintDebug
You don’t need to install Gradle separately because the repository includes the Gradle wrapper.
After Gradle reports BUILD SUCCESSFUL, install the APK and start Image Analysis:
adb install -r app/build/outputs/apk/debug/app-debug.apk
adb shell am start -n com.arm.learningpath.imagetoimage/.ui.MainActivity
adb install -r app\build\outputs\apk\debug\app-debug.apk
adb shell am start -n com.arm.learningpath.imagetoimage/.ui.MainActivity
Select Depth Anything V2 Small INT8 in the model menu. Its status reports placeholder bundled because the real .pte file hasn’t been copied to the phone.
What you’ve accomplished and what’s next
You’ve built, tested, and installed the application with its model-specific Depth Anything adapter.
Next, you’ll download and stage the model artifact.