tools/accuracy_checker/accuracy_checker/adapters/reidentification.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 numpy as np
  18
  19 from ..adapters import Adapter
  20 from ..representation import ReIdentificationPrediction
  21
  22
  23 class ReidAdapter(Adapter):
  24     """
  25     Class for converting output of Reid model to ReIdentificationPrediction representation
  26     """
  27     __provider__ = 'reid'
  28
  29     def configure(self):
  30         """
  31         Specifies parameters of config entry
  32         """
  33         self.grn_workaround = self.launcher_config.get("grn_workaround", True)
  34
  35     def process(self, raw, identifiers=None, frame_meta=None):
  36         """
  37         Args:
  38             identifiers: list of input data identifiers
  39             raw: output of model
  40         Returns:
  41             list of ReIdentificationPrediction objects
  42         """
  43         prediction = self._extract_predictions(raw, frame_meta)[self.output_blob]
  44
  45         if self.grn_workaround:
  46             # workaround: GRN layer
  47             prediction = self._grn_layer(prediction)
  48
  49         return [ReIdentificationPrediction(identifier, embedding.reshape(-1))
  50                 for identifier, embedding in zip(identifiers, prediction)]
  51
  52     @staticmethod
  53     def _grn_layer(prediction):
  54         GRN_BIAS = 0.000001
  55         sum_ = np.sum(prediction ** 2, axis=1)
  56         prediction = prediction / np.sqrt(sum_[:, np.newaxis] + GRN_BIAS)
  57
  58         return prediction