coco decoder

Author: agyurjin

superannotate/input_converters/converters/coco_converters/coco_to_sa_pixel.py

@@ -103,28 +103,51 @@ def coco_instance_segmentation_to_sa_pixel(coco_path, images_path):
     hexcolors = blue_color_generator(len(coco_json['annotations']))
     for i, annot in enumerate(coco_json['annotations']):
         if str(annot['image_id']) not in images_dict:
-            print('check')
             continue
 
         hexcolor = hexcolors[images_dict[str(annot['image_id'])]['segments_num']
                             ]
         color = hex_to_rgb(hexcolor)
         images_dict[str(annot['image_id'])]['segments_num'] += 1
 
-        if isinstance(annot['segmentation'], dict):
-            annot['segmentation'] = _rle_to_polygon(coco_path, annot)
-
-        segment = annot['segmentation'][0]
         H, W, C = images_dict[str(annot['image_id'])]['mask'].shape
-        bitmask = np.zeros((H, W)).astype(np.uint8)
-        pts = np.array(
-            [segment[2 * i:2 * (i + 1)] for i in range(len(segment) // 2)],
-            dtype=np.int32
-        )
+        if isinstance(annot['segmentation'], dict):
+            if isinstance(annot['segmentation']['counts'], list):
+                annot['segmentation'] = _rle_to_polygon(coco_path, annot)
+                for segment in annot['segmentation']:
+                    bitmask = np.zeros((H, W)).astype(np.uint8)
+                    pts = np.array(
+                        [
+                            segment[2 * i:2 * (i + 1)]
+                            for i in range(len(segment) // 2)
+                        ],
+                        dtype=np.int32
+                    )
+
+                    cv2.fillPoly(bitmask, [pts], 1)
+                    images_dict[str(
+                        annot['image_id']
+                    )]['mask'][bitmask == 1] = list(color)[::-1] + [255]
+            else:
+                images_dict[str(
+                    annot['image_id']
+                )]['mask'][maskUtils.decode(annot['segmentation']) == 1
+                          ] = list(color)[::-1] + [255]
+        else:
+            for segment in annot['segmentation']:
+                bitmask = np.zeros((H, W)).astype(np.uint8)
+                pts = np.array(
+                    [
+                        segment[2 * i:2 * (i + 1)]
+                        for i in range(len(segment) // 2)
+                    ],
+                    dtype=np.int32
+                )
 
-        cv2.fillPoly(bitmask, [pts], 1)
-        images_dict[str(annot['image_id']
-                       )]['mask'][bitmask == 1] = list(color)[::-1] + [255]
+                cv2.fillPoly(bitmask, [pts], 1)
+                images_dict[str(
+                    annot['image_id']
+                )]['mask'][bitmask == 1] = list(color)[::-1] + [255]
 
         sa_obj = {
             "classId": annot['category_id'],

superannotate/input_converters/converters/coco_converters/coco_to_sa_vector.py

@@ -11,6 +11,7 @@
 from pycocotools.coco import COCO
 from panopticapi.utils import id2rgb
 from tqdm import tqdm
+import pycocotools.mask as maskUtils
 
 
 # Returns unique values of list. Values can be dicts or lists!
@@ -44,16 +45,40 @@ def coco_instance_segmentation_to_sa_vector(coco_path, images_path):
     for cat in coco_json['categories']:
         cat_id_to_cat[cat['id']] = cat
 
+    instance_groups = {}
+    for annot in coco_json['annotations']:
+        if 'id' in annot:
+            if annot['id'] in instance_groups:
+                instance_groups[annot['id']] += 1
+            else:
+                instance_groups[annot['id']] = 1
+
     for annot in tqdm(coco_json['annotations'], "Converting annotations"):
-        if 'iscrowd' in annot and annot['iscrowd'] == 1:
-            try:
+        if isinstance(annot['segmentation'], dict):
+            if isinstance(annot['segmentation']['counts'], list):
                 annot['segmentation'] = _rle_to_polygon(coco_path, annot)
-            except IndexError:
-                print("List index out of range")
+            else:
+                binary_mask = maskUtils.decode(annot['segmentation'])
+                contours, _ = cv2.findContours(
+                    binary_mask.astype(np.uint8), cv2.RETR_EXTERNAL,
+                    cv2.CHAIN_APPROX_SIMPLE
+                )
+                segments = []
+                for contour in contours:
+                    contour = contour.flatten().tolist()
+                    if len(contour) > 4:
+                        segments.append(contour)
+                annot['segmentation'] = segments
 
         cat = cat_id_to_cat[annot['category_id']]
-        sa_polygon_loader = [
-            {
+        sa_polygon_loader = []
+        groupid = 0
+        for polygon in annot['segmentation']:
+            if 'id' in annot:
+                if annot['id'] in instance_groups and instance_groups[
+                    annot['id']] > 1:
+                    groupid = annot['id']
+            sa_obj = {
                 'type': 'polygon',
                 'points': polygon,
                 'className': cat['name'],
@@ -62,10 +87,10 @@ def coco_instance_segmentation_to_sa_vector(coco_path, images_path):
                 'probability': 100,
                 'locked': False,
                 'visible': True,
-                'groupId': annot['id'],
+                'groupId': groupid,
                 'imageId': annot['image_id']
-            } for polygon in annot['segmentation']
-        ]
+            }
+            sa_polygon_loader.append(sa_obj)
 
         for polygon in sa_polygon_loader:
             if polygon['imageId'] not in image_id_to_annotations:
@@ -79,7 +104,7 @@ def coco_instance_segmentation_to_sa_vector(coco_path, images_path):
             continue
         f_loader = image_id_to_annotations[img['id']]
         if 'file_name' in img:
-            image_path = img['file_name']
+            image_path = os.path.basename(img['file_name'])
         else:
             image_path = img['coco_url'].split('/')[-1]
         file_name = image_path + "___objects.json"
@@ -133,7 +158,7 @@ def coco_object_detection_to_sa_vector(coco_path, images_path):
             continue
         f_loader = image_id_to_annotations[img['id']]
         if 'file_name' in img:
-            image_path = img['file_name']
+            image_path = os.path.basename(img['file_name'])
         else:
             image_path = img['coco_url'].split('/')[-1]
         file_name = image_path + "___objects.json"
@@ -239,6 +264,6 @@ def coco_keypoint_detection_to_sa_vector(coco_path, images_path):
             for img_id, img_data in loader:
                 if img['id'] == img_id:
                     f_loader.append(img_data)
-            file_name = img['file_name'] + "___objects.json"
+            file_name = os.path.basename(img['file_name']) + "___objects.json"
             sa_jsons[file_name] = f_loader
     return sa_jsons

tests/converter_test/COCO/input/fromSuperAnnotate/cats_dogs_desktop/annotations.json

@@ -0,0 +1 @@
+{"dog3.jpg":[{"type":"bbox","classId":1,"probability":100,"points":{"x1":192.5,"x2":554.2,"y1":65.6,"y2":681},"groupId":0,"pointLabels":{},"locked":false,"visible":true,"attributes":[]},{"type":"meta","name":"lastAction","timestamp":1605698344455}],"dog2.jpg":[{"type":"bbox","classId":1,"probability":100,"points":{"x1":119.55786895751953,"x2":407.70001220703125,"y1":28.329744338989258,"y2":458.6000061035156},"groupId":0,"pointLabels":{},"locked":false,"visible":true,"attributes":[]},{"type":"meta","name":"lastAction","timestamp":1605698356828}],"dog1.jpg":[{"type":"bbox","classId":1,"probability":100,"points":{"x1":181.88186645507812,"x2":653.2000122070312,"y1":83.06184387207031,"y2":462.3999938964844},"groupId":0,"pointLabels":{},"locked":false,"visible":true,"attributes":[]},{"type":"meta","name":"lastAction","timestamp":1605698370093}],"cat3.jpg":[{"type":"bbox","classId":2,"probability":100,"points":{"x1":0,"x2":529.7525634765625,"y1":30.130569458007812,"y2":862.4000244140625},"groupId":0,"pointLabels":{},"locked":false,"visible":true,"attributes":[]},{"type":"meta","name":"lastAction","timestamp":1605698384506}],"cat2.jpg":[{"type":"bbox","classId":2,"probability":100,"points":{"x1":254.9,"x2":1043.2,"y1":101.9,"y2":732},"groupId":0,"pointLabels":{},"locked":false,"visible":true,"attributes":[]},{"type":"meta","name":"lastAction","timestamp":1605698428282}],"cat1.jpg":[{"type":"bbox","classId":2,"probability":100,"points":{"x1":483.3,"x2":1547.9,"y1":0,"y2":1924.3},"groupId":0,"pointLabels":{},"locked":false,"visible":true,"attributes":[]},{"type":"meta","name":"lastAction","timestamp":1605698430409}]}

tests/converter_test/COCO/input/fromSuperAnnotate/cats_dogs_desktop/classes.json

@@ -0,0 +1 @@
+[{"attribute_groups":[],"color":"#32a852","id":1,"name":"Dog","opened":true},{"attribute_groups":[],"color":"#6a010f","id":2,"name":"Cat","opened":true}]

tests/converter_test/COCO/input/fromSuperAnnotate/cats_dogs_desktop/config.json

@@ -0,0 +1 @@
+{"pathSeparator":"/","os":"linux"}