add script for generate ground plane line annotations

script/generate_road_line_annotation.py

@@ -0,0 +1,46 @@
+# Auto generate roadline reference to json annotation
+import pydatatool as pdt
+from pycocotools.coco import COCO
+import numpy as np
+from collections import defaultdict
+import os.path
+import logging
+
+logging.basicConfig(level=logging.DEBUG)
+
+json_file = '../output/json/scut_train_10x.json'
+
+scut_train_json = pdt.load_json(json_file)
+
+scut_train = COCO(json_file)
+
+for index, img in enumerate(scut_train_json['images']):
+    anns = scut_train.imgToAnns[img['id']]
+    if len(anns):
+        y_centers = [ann['bbox'][1]+ann['bbox'][3]/2 for ann in anns]
+    else:
+        y_centers = -1
+    # simple average center
+    # TODO(xzw): more smooth raodline
+    roadline = np.mean(y_centers)
+
+    scut_train_json['images'][index]['roadline'] = roadline
+
+    logging.debug("Image ID:{} RoadLine:{}".format(img['id'],roadline))
+
+y_centers = defaultdict(list)
+for img in scut_train_json['images']:
+    y_centers[img['id']] = img['roadline']
+logging.debug("Done.")
+
+output_name,ext = os.path.splitext(json_file)
+output_name = output_name + '_roadline' + ext
+logging.debug("Save json to {}.".format(output_name))
+pdt.save_json(scut_train_json,output_name)
+
+output_name,ext = os.path.splitext(json_file)
+output_name = output_name + '_roadline_only' + ext
+logging.debug("Save roadline json to {}.".format(output_name))
+pdt.save_json(y_centers,output_name)
+
+logging.debug("Done.")