Add 'load_label' parameter for image classification models (#185)

* add 'load_label' parameter for image classification models

* move load_label flag to initializer

models/image_classification_mobilenet/demo.py

@@ -31,13 +31,16 @@ parser.add_argument('--backend_target', '-bt', type=int, default=0,
                         {:d}: TIM-VX + NPU,
                         {:d}: CANN + NPU
                     '''.format(*[x for x in range(len(backend_target_pairs))]))
+parser.add_argument('--top_k', type=int, default=1,
+                    help='Usage: Get top k predictions.')
 args = parser.parse_args()
 
 if __name__ == '__main__':
     backend_id = backend_target_pairs[args.backend_target][0]
     target_id = backend_target_pairs[args.backend_target][1]
+    top_k = args.top_k
     # Instantiate MobileNet
-    model = MobileNet(modelPath=args.model, backendId=backend_id, targetId=target_id)
+    model = MobileNet(modelPath=args.model, topK=top_k, backendId=backend_id, targetId=target_id)
 
     # Read image and get a 224x224 crop from a 256x256 resized
     image = cv.imread(args.input)

models/image_classification_mobilenet/mobilenet.py

@@ -6,10 +6,11 @@ class MobileNet:
     Works with MobileNet V1 & V2.
     '''
 
-    def __init__(self, modelPath, topK=1, backendId=0, targetId=0):
+    def __init__(self, modelPath, topK=1, loadLabel=True, backendId=0, targetId=0):
         self.model_path = modelPath
         assert topK >= 1
         self.top_k = topK
+        self.load_label = loadLabel
         self.backend_id = backendId
         self.target_id = targetId
 
@@ -64,7 +65,7 @@ class MobileNet:
         for o in output_blob:
             class_id_list = o.argsort()[::-1][:self.top_k]
             batched_class_id_list.append(class_id_list)
-        if len(self._labels) > 0:
+        if len(self._labels) > 0 and self.load_label:
             batched_predicted_labels = []
             for class_id_list in batched_class_id_list:
                 predicted_labels = []

models/image_classification_ppresnet/demo.py

@@ -37,13 +37,16 @@ parser.add_argument('--backend_target', '-bt', type=int, default=0,
                         {:d}: TIM-VX + NPU,
                         {:d}: CANN + NPU
                     '''.format(*[x for x in range(len(backend_target_pairs))]))
+parser.add_argument('--top_k', type=int, default=1,
+                    help='Usage: Get top k predictions.')
 args = parser.parse_args()
 
 if __name__ == '__main__':
     backend_id = backend_target_pairs[args.backend_target][0]
     target_id = backend_target_pairs[args.backend_target][1]
+    top_k = args.top_k
     # Instantiate ResNet
-    model = PPResNet(modelPath=args.model, backendId=backend_id, targetId=target_id)
+    model = PPResNet(modelPath=args.model, topK=top_k, backendId=backend_id, targetId=target_id)
 
     # Read image and get a 224x224 crop from a 256x256 resized
     image = cv.imread(args.input)

models/image_classification_ppresnet/ppresnet.py

@@ -9,10 +9,11 @@ import numpy as np
 import cv2 as cv
 
 class PPResNet:
-    def __init__(self, modelPath, topK=1, backendId=0, targetId=0):
+    def __init__(self, modelPath, topK=1, loadLabel=True, backendId=0, targetId=0):
         self._modelPath = modelPath
         assert topK >= 1
         self._topK = topK
+        self._load_label = loadLabel
         self._backendId = backendId
         self._targetId = targetId
 
@@ -69,7 +70,7 @@ class PPResNet:
         for ob in outputBlob:
             class_id_list = ob.argsort()[::-1][:self._topK]
             batched_class_id_list.append(class_id_list)
-        if len(self._labels) > 0:
+        if len(self._labels) > 0 and self._load_label:
             batched_predicted_labels = []
             for class_id_list in batched_class_id_list:
                 predicted_labels = []

tools/eval/eval.py

@@ -25,32 +25,38 @@ models = dict(
         name="MobileNet",
         topic="image_classification",
         modelPath=os.path.join(root_dir, "models/image_classification_mobilenet/image_classification_mobilenetv1_2022apr.onnx"),
-        topK=5),
+        topK=5,
+        loadLabel=False),
     mobilenetv1_q=dict(
         name="MobileNet",
         topic="image_classification",
         modelPath=os.path.join(root_dir, "models/image_classification_mobilenet/image_classification_mobilenetv1_2022apr_int8.onnx"),
-        topK=5),
+        topK=5,
+        loadLabel=False),
     mobilenetv2=dict(
         name="MobileNet",
         topic="image_classification",
         modelPath=os.path.join(root_dir, "models/image_classification_mobilenet/image_classification_mobilenetv2_2022apr.onnx"),
-        topK=5),
+        topK=5,
+        loadLabel=False),
     mobilenetv2_q=dict(
         name="MobileNet",
         topic="image_classification",
         modelPath=os.path.join(root_dir, "models/image_classification_mobilenet/image_classification_mobilenetv2_2022apr_int8.onnx"),
-        topK=5),
+        topK=5,
+        loadLabel=False),
     ppresnet=dict(
         name="PPResNet",
         topic="image_classification",
         modelPath=os.path.join(root_dir, "models/image_classification_ppresnet/image_classification_ppresnet50_2022jan.onnx"),
-        topK=5),
+        topK=5,
+        loadLabel=False),
     ppresnet_q=dict(
         name="PPResNet",
         topic="image_classification",
         modelPath=os.path.join(root_dir, "models/image_classification_ppresnet/image_classification_ppresnet50_2022jan_int8.onnx"),
-        topK=5),
+        topK=5,
+        loadLabel=False),
     yunet=dict(
         name="YuNet",
         topic="face_detection",