A Novel Memory-Scheduling Strategy for Large Convolutional Neural Network on Memory-Limited Devices

Table 5

Memory usage in each layer for VGG-S.

AlexNet layer	Memory limit(MB)
	200	100	50	10	5
1	0.63	0.63	0.63	0.63	0.60
2	4.97	4.97	4.97	4.97	4.97
3	6.74	6.74	6.74	5.56	4.39
4	8.78	8.78	8.78	8.78	4.39
5	4.88	4.88	4.88	4.88	4.88
6	1.52	1.52	1.52	1.52	3.86
7	5.53	5.53	5.53	5.53	3.28
8	1.29	1.29	1.29	1.29	1.29
9	0.77	0.77	0.77	0.77	0.77
10	9.52	9.52	9.52	5.02	3.52
11	1.03	1.03	1.03	1.03	1.03
12	9.52	9.52	9.52	5.02	3.52
13	1.03	1.03	1.03	1.03	1.03
14	144.52	96.52	0.52	0.52	0.52
15	0.57	0.57	0.57	0.57	0.57
16	0.06	0.06	48.06	9.99	4.94
17	64.02	64.02	32.01	0.01	0.01
18	0.01	0.01	0.01	9.15	4.93
19	15.63	15.63	15.63	7.82	3.91
20	0.01	0.01	0.01	0.01	0.01