机器学习
深度学习
长按二维码关注
Type / Stride | Filter Shape | Input Size |
---|---|---|
Conv / s2 | 3*3*3*32 | 224*224*3 |
Conv dw / s1 | 3*3*32 dw | 112*112*32 |
Conv / s1 | 1*1*32*64 | 112*112*32 |
Conv dw / s2 | 3*3*64 dw | 112*112*64 |
Conv / s1 | 1*1*64*128 | 56*56*64 |
Conv dw / s1 | 3*3*128 dw | 56*56*128 |
Conv / s1 | 1*1*128*128 | 56*56*128 |
Conv dw / s2 | 3*3*128 dw | 56*56*128 |
Conv / s1 | 1*1*128*256 | 28*28*128 |
Conv dw / s1 | 3*3*256 dw | 28*28*256 |
Conv / s1 | 1*1*256*256 | 28*28*256 |
Conv dw / s2 | 3*3*256 dw | 28*28*256 |
Conv / s1 | 1*1*256*512 | 14*14*256 |
5 * Conv dw / s1 Conv / s1 | 3*3*512 dw 1*1*512*512 | 14*14*51214*14*512 |
Conv dw / s2 | 3*3*512 dw | 14*14*512 |
Conv / s1 | 1*1*512*1024 | 7*7*1024 |
Conv dw / s2 | 3*3*1024 dw | 7*7*1024 |
Conv / s1 | 1*1*1024*1024 | 7*7*1024 |
Avg Pool / s1 | Pool 7*7 | 7*7*1024 |
FC | 1024*1000 | 1*1*1024 |
Softmax | Classfier | 1*1*1000 |
Layer | Output size | KSize | Stride | Repeat | Ouput channels(ggg groups) ggg=1 ggg=2 ggg=3 ggg=4 ggg=8 |
---|---|---|---|---|---|
Image | 224*224 | 3 3 3 3 3 | |||
Conv1 MaxPool | 112*112 56*56 | 21 | 13 | 24 24 24 24 24 | |
Stage2 | 28*2828*28 | 21 | 13 | 144 200 240 272 384 | |
Stage3 | 14*1414*14 | 21 | 13 | 288 400 480 544 768 | |
Stage4 | 7*77*7 | 21 | 17 | 576 800 960 1088 1536 | |
GlobalPool | 1*1 | 7*7 | |||
FC | 1000 1000 1000 1000 1000 | ||||
Complexity | 143M 140M 137M 133M 137M |
Input | Operator | Output |
---|---|---|
h∗w∗kh*w*kh∗w∗k | 1*1 Conv, ReLU6 | h∗w∗tkh*w*tkh∗w∗tk |
h∗w∗tkh*w*tkh∗w∗tk | 3*3 dw Conv st=sss, ReLU6 | hs∗ws∗tk\frac{h}{s}*\frac{w}{s}*tksh∗sw∗tk |
hs∗ws∗tk\frac{h}{s}*\frac{w}{s}*tksh∗sw∗tk | Linear 1*1 Conv | h∗w∗k∗h*w*k^*h∗w∗k∗ |
Input | Operator | ttt | ccc | nnn | sss |
---|---|---|---|---|---|
224*224*3 | Conv 3*3 | - | 32 | 1 | 2 |
112*112*32 | bottleneck | 1 | 16 | 1 | 1 |
112*112*16 | bottleneck | 6 | 24 | 2 | 2 |
56*56*24 | bottleneck | 6 | 32 | 3 | 2 |
28*28*32 | bottleneck | 6 | 64 | 4 | 2 |
14*14*64 | bottleneck | 6 | 96 | 3 | 1 |
14*14*96 | bottleneck | 6 | 160 | 3 | 2 |
7*7*160 | bottleneck | 6 | 320 | 1 | 1 |
7*7*320 | Conv 1*1 | - | 1280 | 1 | 1 |
7*7*1280 | avgpool 7*7 | - | - | 1 | - |
1*1*1280 | conv 1*1 | - | k | - |
Layer | Output Size | KSize | Stride | Repeat | Output channels 0.5* 1.0* 1.5* 2.0* |
---|---|---|---|---|---|
Image | 224*224 | 3 3 3 3 | |||
Conv1MaxPool | 112*11256*56 | 3*3 | 22 | 1 | 24 24 24 24 |
Stage2 | 28*2828*28 | 21 | 13 | 48 116 176 244 | |
Stage3 | 14*1414*14 | 21 | 17 | 96 232 352 488 | |
Stage4 | 7*77*7 | 21 | 13 | 192 464 704 976 | |
Conv5 | 7*7 | 1*1 | 1024 1024 1024 2048 | ||
GlobalPool | 1*1 | 7*7 | |||
FC | 1000 1000 1000 1000 | ||||
FLOPS | 41M 146M 299M 591M | ||||
# of Weights | 1.4M 2.3M 3.5M 7.4M |
Model | TOP-1 Accuary | Million Muli-Adds | Million Parameters |
---|---|---|---|
GoogleNet | 69.8% | 1550 | 6.8 |
VGG 16 | 71.5% | 15300 | 138 |
Inception V3 | 84% | 5000 | 23.2 |
1.0 MobileNet-224 | 70.6% | 569 | 4.2 |
ShuffleNet 1.5*(g=3) | 71.5% | 292 | 3.4 |
ShuffleNet 2*(g=3) | 73.7% | 524 | 5.4 |
MobileNet v2 | 72.0% | 300 | 3.4 |
MobileNet v2(1.4) | 74.7% | 585 | 6.9 |
ShuffleNet v2 2* | 74.9% | 591 | 6.7 |
Model | MFLOPS | TOP1-err | GPU Speed(Batches\sec.) |
---|---|---|---|
ShuffleNet v2 0.5* | 41 | 39.7 | 417 |
ShuffleNet v1 (g=3) 0.5* | 38 | 55.1 | 351 |
0.25 MobileNet v1 | 41 | 49.7 | 502 |
0.15 MobileNet v2 | 39 | 55.1 | 351 |
0.4 MobileNet v2 | 43 | 43.4 | 333 |
ShuffleNet v2 1.0* | 146 | 30.6 | 341 |
ShuffleNet v1(g=3) 1.0* | 140 | 32.6 | 213 |
0.5 MobileNet v1 | 149 | 36.3 | 382 |
0.75 MobileNet v2 | 145 | 32.1 | 235 |
0.6 MobileNet v2 | 141 | 33.3 | 249 |
ShuffleNet v2 1.5* | 299 | 27.4 | 255 |
ShuffleNet v1(g=3) 1.5* | 292 | 28.5 | 164 |
0.75 MobileNet v1 | 325 | 31.6 | 314 |
1.0 MobileNet v2 | 300 | 28 | 180 |
本文分享自 机器学习与python集中营 微信公众号,前往查看
如有侵权,请联系 cloudcommunity@tencent.com 删除。
本文参与 腾讯云自媒体分享计划 ,欢迎热爱写作的你一起参与!