README
Last updated
Last updated
This folder contains building code for MobileNetV2, based on MobileNetV2: Inverted Residuals and Linear Bottlenecks
This is the timing of MobileNetV1 vs MobileNetV2 using TF-Lite on the large core of Pixel 1 phone.
MACs, also sometimes known as MADDs - the number of multiply-accumulates needed to compute an inference on a single image is a common metric to measure the efficiency of the model.
Below is the graph comparing V2 vs a few selected networks. The size of each blob represents the number of parameters. Note for ShuffleNet there are no published size numbers. We estimate it to be comparable to MobileNetV2 numbers.
Classification Checkpoint
MACs (M)
Parameters (M)
Top 1 Accuracy
Top 5 Accuracy
Mobile CPU (ms) Pixel 1
582
6.06
75.0
92.5
138.0
509
5.34
74.4
92.1
123.0
300
3.47
71.8
91.0
73.8
221
3.47
70.7
90.1
55.1
154
3.47
68.8
89.0
40.2
99
3.47
65.3
86.9
27.6
56
3.47
60.3
83.2
17.6
209
2.61
69.8
89.6
55.8
153
2.61
68.7
88.9
41.6
107
2.61
66.4
87.3
30.4
69
2.61
63.2
85.3
21.9
39
2.61
58.8
81.6
14.2
97
1.95
65.4
86.4
28.7
71
1.95
63.9
85.4
21.1
50
1.95
61.0
83.2
14.9
32
1.95
57.7
80.8
9.9
18
1.95
51.2
75.8
6.4
59
1.66
60.3
82.9
19.7
43
1.66
58.2
81.2
14.6
30
1.66
55.7
79.1
10.5
20
1.66
50.8
75.0
6.9
11
1.66
45.5
70.4
4.5
See this ipython notebook or open and run the network directly in Colaboratory.
