Added ability to use custom labels in conll_ner converter (#3089)

Author: Anna Grebneva

tools/accuracy_checker/openvino/tools/accuracy_checker/annotation_converters/README.md

@@ -707,12 +707,17 @@ The main difference between this converter and `super_resolution` in data organi
 * `see_in_the_dark` - converts See-in-the-Dark dataset described in the [paper](https://cchen156.github.io/paper/18CVPR_SID.pdf) to `ImageProcessingAnnotation`.
   * `annotation_file` - path to image pairs file in txt format.
 * `conll_ner` - converts CONLL 2003 dataset for Named Entity Recognition to `BERTNamedEntityRecognitionAnnotation`.
-  * `annotation_file` - annotation file in txt forma
+  * `annotation_file` - annotation file in txt format.
   * `vocab_file` - vocab file for word piece tokenization.
   * `lower_case` - converts all tokens to lower case during tokenization (Optional, default `False`).
   * `max_length` - maximal input sequence length (Optional, default 128).
   * `pad_input` - allow padding for input sequence if input less that `max_length` (Optional, default `True`).
   * `include_special_token_lables` - allow extension original dataset labels with special token labels (`[CLS'`, `[SEP]`]) (Optional, default `False`).
+  * `labels_file` - path to file with custom labels in json format (Optional).
+  Example of labels_file content:
+    ```json
+    {"labels": ["O", "B-PER", "I-PER", "B-ORG", "I-ORG", "B-LOC", "I-LOC", "B-MISC", "I-MISC"]}
+    ```
 * `tacotron2_data_converter` - converts input data for custom tacotron2 pipeline.
   * `annotation_file` - tsv file with location input data and reference.
 * `noise_suppression_dataset` - converts dataset for audio denoising to `NoiseSuppressionAnnotation`

tools/accuracy_checker/openvino/tools/accuracy_checker/annotation_converters/conll_ner.py

@@ -18,6 +18,7 @@
 from ._nlp_common import WordPieceTokenizer
 from ..config import BoolField, PathField, NumberField
 from ..representation import BERTNamedEntityRecognitionAnnotation
+from ..utils import read_json
 
 
 class CONLLDatasetConverter(FileBasedAnnotationConverter):
@@ -36,7 +37,8 @@ def parameters(cls):
             ),
             'include_special_token_labels': BoolField(
                 optional=True, default=False, description='Should special tokens be included to labels or not'
-            )
+            ),
+            'labels_file': PathField(optional=True, description='Path to file with custom labels in json format')
         })
         return params
 
@@ -47,6 +49,9 @@ def configure(self):
             self.get_value_from_config('vocab_file'),
             lower_case=self.get_value_from_config('lower_case'), max_len=self.get_value_from_config('max_len')
         )
+        labels_file = self.get_value_from_config('labels_file')
+        if labels_file:
+            self.label_list = read_json(labels_file)['labels']
         if self.include_spec:
             self.label_list.extend(['[CLS]', '[SEP]'])
         self.pad = self.get_value_from_config('pad_input')