ABCDEFGHIJKLMNOPQRSTUVWXYZ
1
WebNN ExamplesDirectMLWebGL
2
3
Image Classification
MobileNet v1(TFLite)
17.22+-1.6151.42+-6.81
4
MobileNet v2(TFLite)
19.72+-1.3645.29+-5.78
5
SqueezeNet(TFLite)
16.47+-1.5952.63+-11.00
6
Inception v3(TFLite)
64.54+-2.21189.72+-14.82
7
Inception v4(TFLite)
123.66+-2.09366.40+-6.11
8
Inception Resnet v2(TFLite)
128.02+-2.34325.44+-6.82
9
SqueezeNet(ONNX)
9.87+-0.8645.01+-4.05
10
MobileNet v2(ONNX)
29.35+-1.8454.70+-5.12
11
ResNet50 v1(ONNX)
47.83+-2.29118.13+-10.47
12
ResNet50 v2(ONNX)
72.21+-2.19196.37+-12.68
13
Inception v2(ONNX)
34.83+-2.1979.93+-10.76
14
Object Detection
SSD MobileNet v1(TFLite)
31.30+-1.68116.30+-9.21
15
SSD MobileNet v2(TFLite)
44.89+-1.78116.96+-14.28
16
17
OS: Windows 10 Pro Insider Preview, Version: 1903, OS build: 18975.1000
GPU Driver Version: 26.20.100.6999,
Device: Dell XPS 13,
CPU: Intel i5-8250U,
GPU: Intel UHD Graphics 620,
Memory: 8GB
WebNN examples: https://intel.github.io/webml-polyfill/examples/
The data is average (200 iterations) inference time (1 batch) in milliseconds.
WebGL op is based on TensorFlow.js 1.2.2. DML uses fp16 precision.
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100