添加注释

LPRNet_ORT_infer.py

@@ -27,7 +27,10 @@ def LPRNetPostprocess(infer_res):
     for j in range(infer_res.shape[1]):
         preb_label.append(np.argmax(infer_res[:, j], axis=0))
     no_repeat_blank_label = []
+
+    print(preb_label)
     pre_c = preb_label[0]
+    print(pre_c)
     if pre_c != len(CHARS) - 1:
         no_repeat_blank_label.append(pre_c)
     for c in preb_label:  # dropout repeate label and blank label

LPRNet_migraphx_infer.py

@@ -52,7 +52,8 @@ def LPRNetInference(model_name, imgs):
         model = migraphx.parse_onnx(model_name)
         # migraphx.quantize_fp16(model)
         model.compile(t=migraphx.get_target("gpu"),device_id=0) # device_id: 设置GPU设备，默认为0号设备(>=1.2版本中支持)
-        # migraphx.save(model, 'LPRNet.mxr')
+        path = model_name.split('/')
+        migraphx.save(model, '/'.join((path[:-1], path[-1][-4]+'mxr')))
 
     inputName=model.get_parameter_names()[0]
     inputShape=model.get_parameter_shapes()[inputName].lens()
@@ -64,8 +65,8 @@ def LPRNetInference(model_name, imgs):
     return result
 
 if __name__ == '__main__':
-    # model_name = 'LPRNet.onnx'
-    model_name = 'model/LPRNet.mxr'
+    model_name = 'LPRNet.onnx'
+    # model_name = 'model/LPRNet.mxr'
     image = 'imgs/川JK0707.jpg'
     InferRes = LPRNetInference(model_name, image)
     print(image, 'Inference Result:', InferRes)

README.md

@@ -8,3 +8,9 @@
 
 对于车牌检测，也可以使用图像分割的思想，例如使用UNet语义分割网络，分割出车牌，
 二值化然后查找连通域，计算4个顶点
+
+导出onnx模型：
+python test.py --export_onnx true
+
+推理onnx模型：
+python LPRNet_ORT_infer.py

test.py

@@ -11,7 +11,7 @@ def validation(args):
     model.load_state_dict(torch.load(args.model, map_location=args.device))
     model.to(args.device)
 
-    img = cv2.imread(args.imgpath)
+    img = cv2.imread(args.img)
     height, width, _ = img.shape
     if height != 24 or width != 94:
         img = cv2.resize(img, (94, 24))