Export MobileSAM for Ethos-U85

Prepare the MobileSAM source and checkpoint

Run all commands from the ExecuTorch repository root. Confirm that your Python environment is active and that you’ve sourced the Arm tools’ setup_path.sh file in the current shell, as shown during environment setup.

Prepare the pinned MobileSAM source and checkpoint:

    

        
        
python3 examples/arm/mobilesam_prompt_segmentation_example_ethos_u/model_export/prepare_mobilesam.py

    

The script downloads the pinned MobileSAM revision , applies the patch needed for a configurable image size, and verifies the checkpoint’s SHA-256 checksum. It stores the source in source/ and the checkpoint in mobile_sam.pt, outside the ExecuTorch repository. The cache directory is:

    

        
        
~/.cache/executorch/mobilesam/f706ad9c4eb7f219c00d9050e46328518ffb65d2/

    

Export the ExecuTorch program

Export the example image and positive point prompt for the ethos-u85-256 target:

    

        
        
python3 examples/arm/mobilesam_prompt_segmentation_example_ethos_u/model_export/export_mobilesam.py

    

The exporter loads the prepared checkpoint, calibrates post-training quantization with the example image, checks the quantized mask against the floating-point mask, and lowers the graph to Ethos-U85.

This script uses one fixed configuration: examples/models/dinov2/dog.jpg, positive point (219, 193), input size 448, target ethos-u85-256, and memory mode Dedicated_Sram_384KB. These values are set in the Python source rather than through command-line arguments.

Export requires a host mask intersection over union (IoU) of at least 0.9 and exactly one Ethos-U delegated subgraph. It writes mobilesam.pte to arm_test/mobilesam/export/.

Verify the export artifacts

The files under arm_test/mobilesam/export/ support the remaining steps:

Check that these artifacts exist before building the runner:

    

        
        
export_dir=arm_test/mobilesam/export

for artifact in \
  mobilesam.pte \
  input.bin \
  input.png \
  fp32_mask.png \
  quantized_mask.png \
  metrics.json \
  delegation.txt; do
  test -s "$export_dir/$artifact" || {
    echo "Missing export artifact: $export_dir/$artifact" >&2
    exit 1
  }
done

test -n "$(find "$export_dir/artifacts" -type f -print -quit)" || {
  echo "No TOSA or Vela artifacts found in $export_dir/artifacts" >&2
  exit 1
}

    

The checks complete without output when the artifacts are present. The point prompt is embedded in the .pte, which is compiled into the runner. The image remains a runtime input: the runner reads input.bin through semihosting, which lets the FVP access files on the host.

What you’ve accomplished and what’s next

You’ve prepared MobileSAM and exported a quantized ExecuTorch program for Ethos-U85.

Next, you’ll build the bare-metal application and run it on the Corstone-320 FVP.