tools/accuracy_checker/accuracy_checker/postprocessor/zoom_segmentation_mask.py

   1 """
   2 Copyright (C) 2018-2019 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 .postprocessor import Postprocessor, BasePostprocessorConfig
  20 from ..representation import SegmentationAnnotation, SegmentationPrediction
  21 from ..config import NumberField
  22
  23
  24 class ZoomSegMask(Postprocessor):
  25     """
  26     Zoom probabilities of segmentation prediction.
  27     """
  28
  29     __provider__ = 'zoom_segmentation_mask'
  30
  31     annotation_types = (SegmentationAnnotation, )
  32     prediction_types = (SegmentationPrediction, )
  33
  34     def validate_config(self):
  35         class _ZoomSegMaskConfigValidator(BasePostprocessorConfig):
  36             zoom = NumberField(floats=False, min_value=1)
  37
  38         zoom_segmentation_mask_config_validator = _ZoomSegMaskConfigValidator(
  39             self.__provider__, on_extra_argument=_ZoomSegMaskConfigValidator.ERROR_ON_EXTRA_ARGUMENT
  40         )
  41         zoom_segmentation_mask_config_validator.validate(self.config)
  42
  43     def configure(self):
  44         self.zoom = self.config['zoom']
  45
  46     def process_image(self, annotation, prediction):
  47         for annotation_, prediction_ in zip(annotation, prediction):
  48             height, width = annotation_.mask.shape[:2]
  49             prob = prediction_.mask
  50             zoom_prob = np.zeros((prob.shape[0], height, width), dtype=np.float32)
  51             for c in range(prob.shape[0]):
  52                 for h in range(height):
  53                     for w in range(width):
  54                         r0 = h // self.zoom
  55                         r1 = r0 + 1
  56                         c0 = w // self.zoom
  57                         c1 = c0 + 1
  58                         rt = float(h) / self.zoom - r0
  59                         ct = float(w) / self.zoom - c0
  60                         v0 = rt * prob[c, r1, c0] + (1 - rt) * prob[c, r0, c0]
  61                         v1 = rt * prob[c, r1, c1] + (1 - rt) * prob[c, r0, c1]
  62                         zoom_prob[c, h, w] = (1 - ct) * v0 + ct * v1
  63             prediction_.mask = zoom_prob
  64
  65         return annotation, prediction