add infer.py

Author: remloveh

tests/yamrt/infer.py

@@ -0,0 +1,63 @@
+import onnxruntime
+import numpy as np
+import cv2
+import json
+import os
+import time
+
+providers = ['CPUExecutionProvider']
+session = onnxruntime.InferenceSession('resnet18v1/resnet18v1.onnx', providers=providers)
+input_name = session.get_inputs()[0].name
+output_name = session.get_outputs()[0].name
+
+def load_labels(path):
+    with open(path) as f:
+        data = json.load(f)
+    return np.asarray(data)
+
+def softmax(x):
+    x = x.reshape(-1)
+    e_x = np.exp(x - np.max(x))
+    return e_x / e_x.sum(axis=0)
+
+labels = load_labels('labels.json')
+
+path = '/home/remloveh/.mxnet/datasets/imagenet/val/'
+li = os.listdir(path)
+li.sort()
+
+total = 0
+right = 0
+label_num = 0
+start = time.time()
+for di in li:
+    ll = os.listdir(path + di)
+    for dd in ll:
+        img = cv2.imread(path + di + '/' + dd)
+        img = cv2.resize(img, (224, 224))
+        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+        
+        data = np.array(img).transpose(2, 0, 1)
+        data = data.astype('float32')
+        mean_vec = np.array([0.485, 0.456, 0.406])
+        stddev_vec = np.array([0.229, 0.224, 0.225])
+        norm_data = np.zeros(data.shape).astype('float32')
+        
+        for i in range(data.shape[0]):
+            norm_data[i,:,:] = (data[i,:,:]/255 - mean_vec[i]) / stddev_vec[i]
+        norm_data = norm_data.reshape(1, 3, 224, 224).astype('float32')
+        
+        result = session.run([output_name],{input_name:norm_data})
+        res = softmax(np.array(result)).tolist()
+        idx = np.argmax(res)
+        
+        if idx == label_num:
+            right = right + 1
+        total = total + 1
+    label_num = label_num + 1
+end = time.time()
+
+print('time: ', int(end - start)/ total, 's')
+
+
+print('accuracy: ', right/total)