add remote test for gps

test/models/private_tests/test_gps.py

@@ -0,0 +1,32 @@
+import onnxruntime as ort
+import numpy as np
+import onnx
+from onnx2kerastl import onnx_to_keras
+from keras_data_format_converter import convert_channels_first_to_last
+import tensorflow as tf
+from test.models.private_tests.aws_utils import aws_s3_download
+import pytest
+import onnxruntime as rt
+
+
+@pytest.mark.parametrize('aws_s3_download', [["gps/", "gps/", False]], indirect=True)
+def test_gps(aws_s3_download):
+    onnx_model_path = f'{aws_s3_download}/gps_750_v1.onnx'
+    onnx_model = onnx.load(onnx_model_path)
+    keras_model = onnx_to_keras(onnx_model, ['images', 'gps', 'masks'], name_policy='attach_weights_name',
+                                allow_partial_compilation=False).converted_model
+    final_model = convert_channels_first_to_last(keras_model, should_transform_inputs_and_outputs=True)
+    data = np.random.random((1, 11, 3, 224, 224))
+    gps = np.random.random((1, 10, 2))
+    masks = np.ones((1, 11))
+    masks[:, :8] = 0
+    res = final_model([data.transpose([0, 2, 3, 4, 1]), masks, gps.transpose([0,2,1])])
+    sess = rt.InferenceSession(onnx_model_path)
+    input_name_1 = sess.get_inputs()[0].name
+    input_name_2 = sess.get_inputs()[1].name
+    input_name_3 = sess.get_inputs()[2].name
+    label_name = sess.get_outputs()[0].name
+    pred = sess.run([label_name],
+                    {input_name_1: data.astype(np.float32), input_name_2: gps.astype(np.float32),
+                     input_name_3: masks.astype(np.float32)})
+    assert (pred[0] - res).numpy().__abs__().max() < 2e-5