NXP backend: Enable MobileNetV3 model

backends/nxp/tests/generic_tests/test_aot_example.py

@@ -125,7 +125,7 @@ def _assert_profiling(
 
 
 def test_aot_example__mobilenet_v2():
-    """Test that mobilenet can be lowered to Neutron backend via `aot_neutron_compile.py` and all ops are delegated."""
+    """Test that MobileNetV2 can be lowered to Neutron backend via `aot_neutron_compile.py` and all ops are delegated."""
 
     num_random_samples = 1
 
@@ -154,7 +154,7 @@ def test_aot_example__mobilenet_v2():
 
 
 def test_aot_example__mobilenet_v2__profiling():
-    """Test that mobilenet_v2 can be lowered to Neutron backend via `aot_neutron_compile.py`, all ops are delegated,
+    """Test that MobileNetV2 can be lowered to Neutron backend via `aot_neutron_compile.py`, all ops are delegated,
     the output model is profilable and ETRecord is generated properly."""
 
     num_random_samples = 1

backends/nxp/tests/models/test_mobilenet_v3.py

@@ -0,0 +1,102 @@
+# Copyright 2026 NXP
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+from functools import partial
+
+import numpy as np
+
+# noinspection PyUnusedImports
+import pytest
+import torch
+
+from executorch.backends.nxp.tests.dataset_creator import (
+    FromCalibrationDataDatasetCreator,
+)
+from executorch.backends.nxp.tests.executorch_pipeline import ModelInputSpec
+from executorch.backends.nxp.tests.graph_verifier import (
+    BaseGraphVerifier,
+    NonDelegatedNode,
+)
+from executorch.backends.nxp.tests.model_output_comparator import (
+    NumericalStatsOutputComparator,
+)
+from executorch.backends.nxp.tests.nsys_testing import lower_run_compare
+from executorch.backends.nxp.tests.use_qat import *  # noqa F403
+from executorch.examples.nxp.models.mobilenet_v3 import MobileNetV3
+
+BOUNDS_MSE = {
+    "PTQ": {"channels-last": 7e-03, "channels-first": 8e-03},
+    "QAT": {"channels-last": 6e-03, "channels-first": 3e-03},
+}
+
+
+@pytest.fixture(autouse=True)
+def reseed_model_per_test_run():
+    torch.manual_seed(23)
+    np.random.seed(23)
+
+
+@pytest.mark.parametrize("channels_last", [False, True])
+def test_mobilenet_v3_mse_cpu_vs_npu(
+    mocker,
+    request,
+    channels_last,
+    use_qat,
+):
+    num_samples = 1
+
+    mobilenet_v3 = MobileNetV3(
+        num_samples=num_samples,
+        use_random_dataset=True,
+        balanced_dataset=False,
+    )
+    model = mobilenet_v3.get_eager_model()
+    dataset = mobilenet_v3.dataset
+    labels = mobilenet_v3.labels
+
+    dataset_creator = FromCalibrationDataDatasetCreator(
+        dataset, num_examples=num_samples, idx_to_label=labels
+    )
+
+    input_spec = ModelInputSpec(mobilenet_v3.input_shape)
+    if channels_last:
+        model.to(memory_format=torch.channels_last)
+        input_spec.dim_order = torch.channels_last
+    quant_type_key = "QAT" if use_qat else "PTQ"
+    format_key = "channels-last" if channels_last else "channels-first"
+
+    mse = BOUNDS_MSE[quant_type_key][format_key]
+    comparator = NumericalStatsOutputComparator(
+        max_mse_error=mse, is_classification_task=True
+    )
+    model_verifier = BaseGraphVerifier(
+        1,
+        [
+            NonDelegatedNode("aten_add_tensor", 13),
+            NonDelegatedNode("aten_div_tensor", 9),
+            NonDelegatedNode("aten_clamp_default", 18),
+        ],
+    )
+    train_fn = (
+        partial(mobilenet_v3.train_model_fn, channels_last=channels_last)
+        if use_qat
+        else None
+    )
+
+    ref_input_spec = ModelInputSpec(input_spec.shape)
+    ref_input_spec.dim_order = torch.contiguous_format
+
+    lower_run_compare(
+        model,
+        [input_spec],
+        model_verifier,
+        request,
+        dataset_creator=dataset_creator,
+        output_comparator=comparator,
+        reference_input_spec=[ref_input_spec],
+        mocker=mocker,
+        use_qat=use_qat,
+        train_fn=train_fn,
+    )

examples/nxp/README.md

@@ -23,6 +23,10 @@ $ ./examples/nxp/setup.sh
 ## Supported models
 * CifarNet
 * MobileNetV2
+* MLPerfTinyAnomalyDetection
+* MLPerfTinyImageClassification
+* MLPerfTinyKeywordSpotting
+* MLPerfTinyVisualWakeWords
 
 ## PyTorch Model Delegation to Neutron Backend
 First we will start with an example script converting the model. This example show the CifarNet model preparation. 

examples/nxp/models/mobilenet_v3.py

@@ -0,0 +1,21 @@
+# Copyright 2026 NXP
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+import torch
+
+from executorch.examples.models.mobilenet_v3 import MV3Model
+from executorch.examples.nxp.models.mobilenet_v2 import MobileNetV2
+
+
+class MobileNetV3(MobileNetV2):
+    """
+    MobileNet V3 model. Shares most parameters with MobileNet V2.
+
+    Calibration data for MobileNet V2 and V3 can be generated by backends/nxp/tests/prepare_calibration_data_mobilenet_v2.py
+    from Imagenette dataset, a subset of 10 classes from ImageNet dataset the model is trained on.
+    """
+
+    def _init_eager_model(self) -> torch.nn.Module:
+        return MV3Model().get_eager_model().eval()