coco updated

Author: agyurjin

superannotate/input_converters/conversion.py

@@ -195,6 +195,11 @@ def export_annotation_format(
 
     """
 
+    if isinstance(input_dir, str):
+        input_dir = Path(input_dir)
+    if isinstance(output_dir, str):
+        output_dir = Path(output_dir)
+
     args = Namespace(
         input_dir=input_dir,
         output_dir=output_dir,
@@ -219,6 +224,7 @@ def import_annotation_format(
     project_type="Vector",
     task="object_detection",
     platform="Web",
+    images_root=''
 ):
     """Converts other annotation formats to SuperAnnotate annotation format. Currently available (project_type, task) combinations for converter
     presented below:
@@ -344,6 +350,8 @@ def import_annotation_format(
                  'vector_annotation' can be used to convert all annotations (point, ellipse, circule, cuboid and etc) to SuperAnnotate vector project.
     :param platform: SuperAnnotate has both 'Web' and 'Desktop' platforms. Choose to which platform you want convert. (Default: "Web")
     :type platform: str
+    :param images_root: Additonal path to images directory in input_dir
+    :type platform: str
 
     """
 
@@ -360,6 +368,7 @@ def import_annotation_format(
         project_type=project_type,
         task=task,
         platform=platform,
+        images_root=images_root
     )
 
     _passes_sanity_checks(args)

superannotate/input_converters/converters/coco_converters/coco_converter.py

@@ -96,12 +96,11 @@ def _create_skeleton(self):
         return out_json
 
     def _load_sa_jsons(self):
-        jsons = []
         if self.project_type == 'Pixel':
-            jsons = glob.glob(os.path.join(self.export_root, '*pixel.json'))
+            jsons_gen = self.export_root.glob('*pixel.json')
         elif self.project_type == 'Vector':
-            jsons = glob.glob(os.path.join(self.export_root, '*objects.json'))
-
+            jsons_gen = self.export_root.glob('*objects.json')
+        jsons = [path for path in jsons_gen]
         self.set_num_converted(len(jsons))
         return jsons
 
@@ -113,11 +112,12 @@ def _prepare_single_image_commons_pixel(self, id_, json_path):
             ]
         )
         rm_len = len('___pixel.json')
-        image_path = json_path[:-rm_len
-                              ] + '___lores.jpg'  # maybe not use low res files?
+        image_path = str(
+            json_path
+        )[:-rm_len] + '___lores.jpg'  # maybe not use low res files?
 
-        sa_ann_json = json.load(open(os.path.join(json_path)))
-        sa_bluemask_path = os.path.join(json_path[:-rm_len] + '___save.png')
+        sa_ann_json = json.load(open(json_path))
+        sa_bluemask_path = str(json_path)[:-rm_len] + '___save.png'
 
         image_info = self.__make_image_info(json_path, id_, self.project_type)
 
@@ -144,12 +144,12 @@ def __make_image_info(self, json_path, id_, source_type):
         elif source_type == 'Vector':
             rm_len = len('___objects.json')
 
-        image_path = json_path[:-rm_len]
+        image_path = str(json_path)[:-rm_len]
 
         img_width, img_height = Image.open(image_path).size
         image_info = {
             'id': id_,
-            'file_name': image_path[len(self.output_dir):],
+            'file_name': Path(image_path).name,
             'height': img_height,
             'width': img_width,
             'license': 1

superannotate/input_converters/converters/coco_converters/coco_strategies.py

@@ -4,6 +4,7 @@
 from collections import namedtuple
 from datetime import datetime
 
+from pathlib import Path
 from panopticapi.utils import IdGenerator, id2rgb
 from PIL import Image
 from tqdm import tqdm
@@ -18,7 +19,7 @@
     coco_keypoint_detection_to_sa_vector, coco_object_detection_to_sa_vector
 )
 from .sa_pixel_to_coco import (
-    sa_pixel_to_coco_instance_segmentation, sa_pixel_to_coco_object_detection,
+    sa_pixel_to_coco_instance_segmentation,
     sa_pixel_to_coco_panoptic_segmentation
 )
 from .sa_vector_to_coco import (
@@ -44,34 +45,32 @@ def __setup_conversion_algorithm(self):
     def __str__(self, ):
         return '{} object'.format(self.name)
 
-    def _sa_to_coco_single(self, id_, json_path, id_generator):
+    def _sa_to_coco_single(self, id_, json_path, id_generator, cat_id_map):
 
         image_commons = self._prepare_single_image_commons(id_, json_path)
-        res = self.conversion_algorithm(image_commons, id_generator)
+        res = self.conversion_algorithm(image_commons, id_generator, cat_id_map)
 
         return res
 
     def sa_to_output_format(self):
         out_json = self._create_skeleton()
         out_json['categories'] = self._create_categories(
-            os.path.join(self.export_root, 'classes_mapper.json')
+            self.export_root / 'classes_mapper.json'
         )
 
-        panoptic_root = os.path.join(
-            self.dataset_name, "panoptic_{}".format(self.dataset_name)
-        )
+        cat_id_map = json.load(open(self.export_root / 'classes_mapper.json'))
 
         images = []
         annotations = []
         id_generator = self._make_id_generator()
-        jsons = glob.glob(
-            os.path.join(self.export_root, '*pixel.json'), recursive=True
-        )
+
+        jsons_gen = self.export_root.glob('*pixel.json')
+        jsons = [path for path in jsons_gen]
 
         for id_, json_ in tqdm(enumerate(jsons, 1)):
-            res = self._sa_to_coco_single(id_, json_, id_generator)
+            res = self._sa_to_coco_single(id_, json_, id_generator, cat_id_map)
 
-            panoptic_mask = json_[:-len('___pixel.json')] + '.png'
+            panoptic_mask = str(json_)[:-len('___pixel.json')] + '.png'
 
             Image.fromarray(id2rgb(res[2])).save(panoptic_mask)
 
@@ -88,16 +87,15 @@ def sa_to_output_format(self):
         out_json['images'] = images
         json_data = json.dumps(out_json, indent=4)
         with open(
-            os.path.join(self.output_dir, '{}.json'.format(self.dataset_name)),
-            'w+'
+            self.output_dir / '{}.json'.format(self.dataset_name), 'w+'
         ) as coco_json:
 
             coco_json.write(json_data)
 
         self.set_num_converted(len(jsons))
 
     def to_sa_format(self):
-        json_data = os.path.join(self.export_root, self.dataset_name + ".json")
+        json_data = self.export_root / (self.dataset_name + ".json")
         sa_classes = self._create_sa_classes(json_data)
         sa_jsons = self.conversion_algorithm(json_data, self.output_dir)
         self.dump_output(sa_classes, sa_jsons)
@@ -137,7 +135,7 @@ def __setup_conversion_algorithm(self):
     def __str__(self, ):
         return '{} object'.format(self.name)
 
-    def _sa_to_coco_single(self, id_, json_path, id_generator):
+    def _sa_to_coco_single(self, id_, json_path, id_generator, cat_id_map):
 
         image_commons = self._prepare_single_image_commons(id_, json_path)
         annotations_per_image = []
@@ -166,23 +164,28 @@ def make_annotation(
             return annotation
 
         res = self.conversion_algorithm(
-            make_annotation, image_commons, id_generator
+            make_annotation, image_commons, id_generator, cat_id_map
         )
         return res
 
     def sa_to_output_format(self):
 
         out_json = self._create_skeleton()
         out_json['categories'] = self._create_categories(
-            os.path.join(self.export_root, 'classes_mapper.json')
+            self.export_root / 'classes_mapper.json'
         )
+
+        cat_id_map = json.load(open(self.export_root / 'classes_mapper.json'))
+
         jsons = self._load_sa_jsons()
         images = []
         annotations = []
         id_generator = self._make_id_generator()
         for id_, json_ in tqdm(enumerate(jsons)):
             try:
-                res = self._sa_to_coco_single(id_, json_, id_generator)
+                res = self._sa_to_coco_single(
+                    id_, json_, id_generator, cat_id_map
+                )
             except Exception as e:
                 raise
             images.append(res[0])
@@ -195,16 +198,15 @@ def sa_to_output_format(self):
 
         json_data = json.dumps(out_json, indent=4)
         with open(
-            os.path.join(self.output_dir, '{}.json'.format(self.dataset_name)),
-            'w+'
+            self.output_dir / '{}.json'.format(self.dataset_name), 'w+'
         ) as coco_json:
             coco_json.write(json_data)
         print("NUMBER OF IMAGES FAILED TO CONVERT", self.failed_conversion_cnt)
 
         self.set_num_converted(len(jsons))
 
     def to_sa_format(self):
-        json_data = os.path.join(self.export_root, self.dataset_name + ".json")
+        json_data = self.export_root / (self.dataset_name + ".json")
         sa_classes = self._create_sa_classes(json_data)
         sa_jsons = self.conversion_algorithm(json_data, self.output_dir)
         self.dump_output(sa_classes, sa_jsons)
@@ -232,12 +234,21 @@ def __make_image_info(self, json_path, id_, source_type):
         elif source_type == 'Vector':
             rm_len = len('___objects.json')
 
-        image_path = json_path[:-rm_len]
+        img_width, img_height = 0, 0
+        json_data = json.load(open(json_path))
+        for annot in json_data:
+            if 'type' in annot and annot['type'] == 'meta':
+                img_height = annot['height']
+                img_width = annot['width']
+
+        image_path = str(json_path)[:-rm_len]
+
+        if img_width == 0 and img_height == 0:
+            img_width, img_height = Image.open(image_path).size
 
-        img_width, img_height = Image.open(image_path).size
         image_info = {
             'id': id_,
-            'file_name': image_path[len(self.output_dir):],
+            'file_name': Path(image_path).name,
             'height': img_height,
             'width': img_width,
             'license': 1
@@ -266,15 +277,14 @@ def sa_to_output_format(self):
         json_data = json.dumps(out_json, indent=4)
 
         with open(
-            os.path.join(self.output_dir, '{}.json'.format(self.dataset_name)),
-            'w+'
+            self.output_dir / '{}.json'.format(self.dataset_name), 'w+'
         ) as coco_json:
             coco_json.write(json_data)
 
         self.set_num_converted(len(out_json['images']))
 
     def to_sa_format(self):
-        json_data = os.path.join(self.export_root, self.dataset_name + ".json")
+        json_data = self.export_root / (self.dataset_name + ".json")
         sa_classes = self._create_sa_classes(json_data)
         sa_jsons = self.conversion_algorithm(json_data, self.output_dir)
         self.dump_output(sa_classes, sa_jsons)

superannotate/input_converters/converters/coco_converters/coco_to_sa_pixel.py

@@ -3,6 +3,7 @@
 
 import cv2
 import numpy as np
+from pathlib import Path
 import pycocotools.mask as maskUtils
 from panopticapi.utils import id2rgb
 from tqdm import tqdm
@@ -39,12 +40,12 @@ def coco_panoptic_segmentation_to_sa_pixel(coco_path, images_path):
 
     sa_jsons = {}
     for annotate in tqdm(annotate_list, "Converting"):
-        annot_name = os.path.splitext(annotate["file_name"])[0]
-        img_cv = cv2.imread(os.path.join(images_path, annot_name + ".png"))
+        annot_name = Path(annotate["file_name"]).stem
+        img_cv = cv2.imread(str(images_path / (annot_name + ".png")))
         if img_cv is None:
             print(
                 "'{}' file dosen't exist!".format(
-                    os.path.join(images_path, annot_name + ".png")
+                    images_path / (annot_name + ".png")
                 )
             )
             continue
@@ -74,13 +75,11 @@ def coco_panoptic_segmentation_to_sa_pixel(coco_path, images_path):
             out_json.append(dd)
 
         img = cv2.cvtColor(img.reshape((H, W, C)), cv2.COLOR_RGB2BGR)
-        cv2.imwrite(
-            os.path.join(images_path, annot_name + ".jpg___save.png"), img
-        )
+        cv2.imwrite(str(images_path / (annot_name + ".jpg___save.png")), img)
 
         file_name = annot_name + ".jpg___pixel.json"
         sa_jsons[file_name] = out_json
-        os.remove(os.path.join(images_path, annot_name + ".png"))
+        (images_path / (annot_name + ".png")).unlink()
     return sa_jsons
 
 
@@ -170,7 +169,7 @@ def coco_instance_segmentation_to_sa_pixel(coco_path, images_path):
 
     for id_, value in images_dict.items():
         img = cv2.imwrite(
-            os.path.join(images_path, value['file_name'] + '___save.png'),
+            str(images_path / (value['file_name'] + '___save.png')),
             value['mask']
         )