added test dataset support

auto_training/config_factories/pvt_config_factory.py

@@ -18,6 +18,8 @@ def make_pvt_cfg(data_path: str,
     cfg.train_img_prefix = f"{cfg.data_root}/train/image_2"
     cfg.val_ann_file = f"{cfg.data_root}/val/coco_val.json"
     cfg.val_img_prefix = f"{cfg.data_root}/val/image_2/"
+    cfg.test_ann_file = f"{cfg.data_root}/test/coco_test.json"
+    cfg.test_img_prefix = f"{cfg.data_root}/test/image_2/"
 
     min_res = (input_res[0], input_res[1] * 0.8)
     training_classes = parse_training_data_classes(cfg.train_ann_file)
@@ -127,8 +129,8 @@ def make_pvt_cfg(data_path: str,
             ]),
         test=dict(
             type=cfg.dataset_type,
-            ann_file=cfg.val_ann_file,
-            img_prefix=cfg.val_img_prefix,
+            ann_file=cfg.test_ann_file,
+            img_prefix=cfg.test_img_prefix,
             pipeline=[
                 dict(type='LoadImageFromFile'),
                 dict(

auto_training/prepare.py

@@ -16,23 +16,25 @@ def prepare_folder(image_path, coco_path, coco, mode):
         json.dump(coco, fp)
 
 
-def make_coco_folder(cocos, coco_path, train_image_path, val_image_path):
+def make_coco_folder(cocos, coco_path, train_image_path, val_image_path, test_image_path):
     prepare_folder(train_image_path, coco_path, cocos[0], mode="train")
     prepare_folder(val_image_path, coco_path, cocos[1], mode="val")
+    prepare_folder(test_image_path, coco_path, cocos[2], mode="test")
 
 
 
 def main():
     args = parse_args()
     target_class_map = json.loads(args.target_class_map)
-    cocos = convert_kitti_files(args.kitti_train, args.kitti_val, target_class_map)
-    make_coco_folder(cocos, args.coco_folder, args.kitti_train, args.kitti_val)
+    cocos = convert_kitti_files(args.kitti_train, args.kitti_val, args.kitti_test, target_class_map)
+    make_coco_folder(cocos, args.coco_folder, args.kitti_train, args.kitti_val, args.kitti_test)
 
 
 def parse_args():
     parser = argparse.ArgumentParser(description='Convert kitti to coco dataset')
     parser.add_argument('kitti_train', type=str, help='train input data path, kitti dataset')
     parser.add_argument('kitti_val', type=str, help='val input data path, kitti dataset')
+    parser.add_argument('kitti_test', type=str, help='test input data path, kitti dataset')
     parser.add_argument('coco_folder', type=str, help='output folder')
     parser.add_argument('--target-class-map', type=str, default="{}", help='target class mapping, json strin format. Map to None if class should not be used.')
     return parser.parse_args()

auto_training/utils/kitti_conversion.py

@@ -73,13 +73,13 @@ def read_files_make_dict(path):
     path_dict = {os.path.basename(file): file for file in files}
     return files, path_dict
 
-def convert_kitti_files(train_path, val_path, target_class_map={}):
+def convert_kitti_files(train_path, val_path, test_path, target_class_map={}):
     image_ids = 0
     category_ids = 0
     annot_ids = 0
     categories = {}
     cocos = []
-    for path in train_path, val_path:
+    for path in train_path, val_path, test_path:
         image_dict = {}
         annot_dict = {}
         files, path_dict = read_files_make_dict(path)

custom/inference.py

@@ -61,8 +61,12 @@ def __init__(
 
     def run(self, threshold=0.5):
         results = []
+        num_images = len(os.listdir(self.inf_dir))
+        counter = 0
         for imn in os.listdir(self.inf_dir):
             img = f"{self.inf_dir}/{imn}"
+            if counter % 100 == 0:
+                print(f"{time.strftime('%Y-%m-%d %H:%M:%S')} - mmdet - INFO - Predict({counter}/{num_images})")
             self.fps_logger.start_record()
             result = inference_detector(self.model, img)
             results.append({img: extract_bounding_boxes(result, threshold)})
@@ -74,6 +78,7 @@ def run(self, threshold=0.5):
                     out_file=f"{self.inf_out_dir}/{imn}",
                     score_thr=threshold,
                 )
+            counter += 1
         if self.output_file:
             with open(self.output_file, "w") as f:
                 json.dump(results, f)