add accuracy checker config for aclnet

models/public/aclnet/accuracy-check.yml

@@ -0,0 +1,23 @@
+models:
+  - name: aclnet
+
+    launchers:
+      - framework: dlsdk
+        adapter: classification
+
+    datasets:
+      - name: sound_classification
+
+        preprocessing:
+          - type: clip_audio
+            duration: 1
+            overlap: 50%
+          - type: audio_normalization
+
+        metrics:
+          - type: accuracy
+            name: top@1
+            top_k: 1
+          - type: accuracy
+            name: top@5
+            top_k: 5

tools/accuracy_checker/accuracy_checker/annotation_converters/README.md

@@ -407,6 +407,10 @@ The main difference between this converter and `super_resolution` in data organi
   * `annotation_file` - path to json file containing annotations to the dataset ({index: {path:"...", labels:[...], bbox:[...] (optional), ...})
   * `label_id` - number of label in the annotation file representing spoof/real labels
   * `dataset_meta_file` - path to json file with dataset meta (e.g. label_map)
+* `sound_classification` - converts dataset for sound classification to `ClassificationAnnotation`. The dataset should be represented by directory with input wav files and annotation in 2 column csv format, where first column is audio file name and second is label id from dataset.
+  * `annotation_file` - csv file with selected subset for evaluation, file structure described above.
+  * `audio_dir` - directory with input data, (optional, required only if you want check file existence during annotation conversion).
+
 ## <a name="customizing-dataset-meta"></a>Customizing Dataset Meta
 There are situations when we need customize some default dataset parameters (e.g. replace original dataset label map with own.)
 You are able to overload parameters such as `label_map`, `segmentation_colors`, `background_label` using `dataset_meta_file` argument.

tools/accuracy_checker/accuracy_checker/annotation_converters/__init__.py

@@ -91,6 +91,7 @@
 from .mpii import MPIIDatasetConverter
 from .mapillary_20 import Mapillary20Converter
 from .antispoofing import AntispoofingDatasetConverter
+from .sound_classification_converter import SoundClassificationFormatConverter
 
 __all__ = [
     'BaseFormatConverter',
@@ -168,5 +169,6 @@
     'PlaceRecognitionDatasetConverter',
     'ClutteredMNISTConverter',
     'MPIIDatasetConverter',
-    'AntispoofingDatasetConverter'
+    'AntispoofingDatasetConverter',
+    'SoundClassificationFormatConverter',
 ]

tools/accuracy_checker/accuracy_checker/annotation_converters/sound_classification_converter.py

@@ -0,0 +1,59 @@
+"""
+Copyright (c) 2018-2020 Intel Corporation
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+      http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+"""
+
+from .format_converter import ConverterReturn, BaseFormatConverter
+from ..config import PathField
+from ..representation import ClassificationAnnotation
+from ..utils import read_csv, get_path, check_file_existence
+
+
+class SoundClassificationFormatConverter(BaseFormatConverter):
+    __provider__ = 'sound_classification'
+
+    @classmethod
+    def parameters(cls):
+        parameters = super().parameters()
+        parameters.update({
+            'annotation_file': PathField(description="Path to annotation in cvs format."),
+            'audio_dir': PathField(
+                is_directory=True, optional=True,
+                description='Path to dataset audio files, used only for content existence check'
+            )
+        })
+
+        return parameters
+
+    def configure(self):
+        self.annotation_file = self.get_value_from_config('annotation_file')
+        self.audio_dir = self.get_value_from_config('audio_dir') or self.annotation_file.parent
+
+    def convert(self, check_content=False, progress_callback=None, progress_interval=100, **kwargs):
+        annotation = []
+        content_errors = [] if check_content else None
+        original_annotation = read_csv(get_path(self.annotation_file), fieldnames=['identifier', 'label'])
+        num_iterations = len(original_annotation)
+        for audio_id, audio in enumerate(original_annotation):
+            identifier = audio['identifier']
+            label = int(audio['label'])
+            if check_content:
+                if not check_file_existence(self.audio_dir / identifier):
+                    content_errors.append('{}: does not exist'.format(self.audio_dir / identifier))
+
+            annotation.append(ClassificationAnnotation(identifier, label))
+            if progress_callback is not None and audio_id % progress_interval == 0:
+                progress_callback(audio_id / num_iterations * 100)
+
+        return ConverterReturn(annotation, None, content_errors)

tools/accuracy_checker/configs/aclnet.yml

@@ -0,0 +1 @@
+../../../models/public/aclnet/accuracy-check.yml

tools/accuracy_checker/dataset_definitions.yml

@@ -1055,3 +1055,12 @@ datasets:
       label_id: 43
     annotation: antispoofing.pickle
     dataset_meta: antispoofing.json
+
+  - name: sound_classification
+    data_source: audio_dataset
+    annotation_conversion:
+      converter: sound_classification
+      annotation_file: audio_dataset/validation.csv
+      audio_dir: audio_dataset/data
+    annotation: sound_classification.pickle
+    reader: wav_reader