AC docs: add note about segmentation_masks_source (openvinotoolkit#1243)

* AC: remove deprecated pipelined mode

* AC docs: add mention abot changes in converters behavior

* Revert "AC: remove deprecated pipelined mode"

This reverts commit f93b429.

* Update tools/accuracy_checker/accuracy_checker/annotation_converters/README.md

Co-authored-by: Vladimir Dudnik <vladimir.dudnik@intel.com>

* Update tools/accuracy_checker/accuracy_checker/annotation_converters/README.md

Co-authored-by: Vladimir Dudnik <vladimir.dudnik@intel.com>

Co-authored-by: Vladimir Dudnik <vladimir.dudnik@intel.com>

Author: eaidova

tools/accuracy_checker/README.md

@@ -172,6 +172,7 @@ And optionally:
 - `preprocessing`: list of preprocessing steps applied to input data. If you want calculated metrics to match reported, you must reproduce preprocessing from canonical paper of your topology or ask topology author about required steps.
 - `postprocessing`: list of postprocessing steps.
 - `reader`: approach for data reading. Default reader is `opencv_imread`.
+- `segmentation_masks_source` - path to directory where gt masks for semantic segmentation task stored.
 
 Also it must contain data related to annotation.
 You can convert annotation inplace using:

tools/accuracy_checker/accuracy_checker/annotation_converters/README.md

@@ -86,6 +86,7 @@ Accuracy Checker supports following list of annotation converters and specific f
   * `images_dir` - path to directory with images related to devkit root (default JPEGImages).
   * `mask_dir` - path to directory with ground truth segmentation masks related to devkit root (default SegmentationClass).
   * `dataset_meta_file` - path path to json file with dataset meta (e.g. label_map, color_encoding).Optional, more details in [Customizing dataset meta](#customizing-dataset-meta) section.
+**Note: since OpenVINO 2020.4 converter behaviour changed. `data_source` parameter of dataset should contains directory for images only, if you have segmentation mask in separated location, please use `segmentation_masks_source` for specifying gt masks location.**
 * `mscoco_detection` - converts MS COCO dataset for object detection task to `DetectionAnnotation`.
   * `annotation_file` - path ot annotation file in json format.
   * `has_background` - allows convert dataset with/without adding background_label. Accepted values are True or False. (default is False).
@@ -198,6 +199,7 @@ Accuracy Checker supports following list of annotation converters and specific f
   * `image_postfix` - postfix part for mask file names (optional, default is `.png`).
   * `mask_loader` - the way how GT mask should be loaded. Supported methods: `pillow`, `opencv`, `nifti`, `numpy`, `scipy`.
   * `dataset_meta_file` - path to json file with prepared dataset meta info. It should contains `label_map` key with dictionary in format class_id: class_name and optionally `segmentation_colors` (if your dataset uses color encoding). Segmentation colors is a list of channel-wise values for each class. (e.g. if your dataset has 3 classes in BGR colors, segmentation colors for it will looks like: `[[255, 0, 0], [0, 255, 0], [0, 0, 255]]`). (Optional, you can provide self-created file as `dataset_meta` in your config).
+**Note: since OpenVINO 2020.4 converter behaviour changed. `data_source` parameter of dataset should contains directory for images only, if you have segmentation mask in separated location, please use `segmentation_masks_source` for specifying gt masks location.**
 * `camvid` - converts CamVid dataset format to `SegmentationAnnotation`.
   * `annotation_file` - file in txt format which contains list of validation pairs (`<path_to_image>` `<path_to_annotation>` separated by space)
   * `dataset_meta_file` - path path to json file with dataset meta (e.g. label_map, color_encoding).Optional, more details in [Customizing dataset meta](#customizing-dataset-meta) section.

tools/accuracy_checker/dataset_definitions.yml

@@ -492,7 +492,8 @@ datasets:
     dataset_meta: vggfaces2.json
 
   - name: semantic_segmentation_adas
-    data_source: segmentation
+    data_source: segmentation/images
+    segmentation_masks_source: segmentation/mask_segmentation_adas
     annotation_conversion:
       converter: common_semantic_segmentation
       images_dir: segmentation/images
@@ -517,7 +518,8 @@ datasets:
         dst_width: 2048
 
   - name: road_segmentation
-    data_source: segmentation
+    data_source: segmentation/images
+    segmentation_masks_source: segmentation/mask_road_segmentation
     annotation_conversion:
       converter: common_semantic_segmentation
       images_dir: segmentation/images