model-optimizer/extensions/front/onnx/detection_output_test.py

   1 """
   2  Copyright (c) 2019 Intel Corporation
   3
   4  Licensed under the Apache License, Version 2.0 (the "License");
   5  you may not use this file except in compliance with the License.
   6  You may obtain a copy of the License at
   7
   8       http://www.apache.org/licenses/LICENSE-2.0
   9
  10  Unless required by applicable law or agreed to in writing, software
  11  distributed under the License is distributed on an "AS IS" BASIS,
  12  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  See the License for the specific language governing permissions and
  14  limitations under the License.
  15 """
  16
  17 import onnx
  18 import unittest
  19
  20 import numpy as np
  21
  22 from extensions.front.onnx.detection_output import DetectionOutputFrontExtractor
  23 from extensions.ops.DetectionOutput import DetectionOutput
  24 from mo.ops.op import Op
  25 from mo.utils.unittest.extractors import PB
  26
  27
  28 class TestDetectionOutputExt(unittest.TestCase):
  29     @staticmethod
  30     def _create_do_node(num_classes=0, share_location=0, background_label_id=0,
  31                         code_type="", variance_encoded_in_target=0, keep_top_k=0,
  32                         confidence_threshold=0, nms_threshold=0, top_k=0, eta=0):
  33         pb = onnx.helper.make_node(
  34             'DetectionOutput',
  35             inputs=['x'],
  36             outputs=['y'],
  37             num_classes=num_classes,
  38             share_location=share_location,
  39             background_label_id=background_label_id,
  40             code_type=code_type,
  41             variance_encoded_in_target=variance_encoded_in_target,
  42             keep_top_k=keep_top_k,
  43             confidence_threshold=confidence_threshold,
  44             # nms_param
  45             nms_threshold=nms_threshold,
  46             top_k=top_k,
  47             eta=eta,
  48         )
  49
  50         node = PB({'pb': pb})
  51         return node
  52
  53     @classmethod
  54     def setUpClass(cls):
  55         Op.registered_ops['DetectionOutput'] = DetectionOutput
  56
  57     def test_do_no_pb_no_ml(self):
  58         self.assertRaises(AttributeError, DetectionOutputFrontExtractor.extract, None)
  59
  60     def test_do_ext_ideal_numbers(self):
  61         node = self._create_do_node(num_classes=21, share_location=1,
  62                                     code_type="CENTER_SIZE", keep_top_k=200,
  63                                     confidence_threshold=0.01, nms_threshold=0.45, top_k=400, eta=1.0)
  64
  65         DetectionOutputFrontExtractor.extract(node)
  66
  67         exp_res = {
  68             'op': 'DetectionOutput',
  69             'type': 'DetectionOutput',
  70             'num_classes': 21,
  71             'share_location': 1,
  72             'background_label_id': 0,
  73             'code_type': "caffe.PriorBoxParameter.CENTER_SIZE",
  74             'variance_encoded_in_target': 0,
  75             'keep_top_k': 200,
  76             'confidence_threshold': 0.01,
  77             'visualize_threshold': 0.6,
  78             # nms_param
  79             'nms_threshold': 0.45,
  80             'top_k': 400,
  81             'eta': 1.0,
  82             # ONNX have not such parameters
  83             # save_output_param.resize_param
  84             'prob': 0,
  85             'resize_mode': "",
  86             'height': 0,
  87             'width': 0,
  88             'height_scale': 0,
  89             'width_scale': 0,
  90             'pad_mode': "",
  91             'pad_value': "",
  92             'interp_mode': "",
  93             'input_width': 1,
  94             'input_height': 1,
  95             'normalized': 1,
  96         }
  97
  98         for key in exp_res.keys():
  99             if key in ['confidence_threshold', 'visualise_threshold', 'nms_threshold', 'eta']:
 100                 np.testing.assert_almost_equal(node[key], exp_res[key])
 101             else:
 102                 self.assertEqual(node[key], exp_res[key])