What you will learn

By the end of this module, you will be able to:

  • export and lower an FP32 segmentation model to ExecuTorch with XNNPACK

  • validate local output using mask, overlay, and transparent-cutout artifacts

  • quantize the model to INT8 and compare artifact size and cutout quality

  • package the INT8 .pte, labels, and model specification into an Android application

  • explain how Android assets, app-private storage, JNI, the ExecuTorch AAR, and Java API fit together

  • interpret preprocessing, forward, postprocessing, and total application time

Previous page Complete and continue