在80分基础上，想起自己的第一个pytorch程序训练cifar10，残差中用了两个3*3的卷积核，
所以把darknet的先降后升的两个卷积核替换！
这不，训练拿下81分！
learn rate：0.0001
轮次：18
rb均值: 0.7696979046,rb方差:5.313940525055
rb均值: 3.5271227360,rb方差:6.893173217773
rb均值: -3.3759846687,rb方差:11.700593948364
rb均值: -1.5381846428,rb方差:7.267606735229
rb均值: -0.0259788577,rb方差:11.587386131287
rb均值: -1.8444420099,rb方差:7.692439556122
rb均值: -1.3937554359,rb方差:13.060191154480
时间: 19921.195313 ms
train Classification result: 94.98% ok (used 49984 images)
时间: 1572.749023 ms
Test Classification result: 79.64% ok (used 9984 images)
learn rate：0.0001
轮次：19
rb均值: 0.7838774323,rb方差:5.403944969177
rb均值: 3.5643568039,rb方差:7.029367923737
rb均值: -3.3800442219,rb方差:11.852703094482
rb均值: -1.5570255518,rb方差:7.349443435669
rb均值: -0.0744066536,rb方差:11.758628845215
rb均值: -1.8746623993,rb方差:7.755856513977
rb均值: -1.4790049791,rb方差:13.170622825623
时间: 19801.056641 ms
train Classification result: 95.62% ok (used 49984 images)
时间: 1564.193970 ms
Test Classification result: 80.01% ok (used 9984 images)
learn rate：0.0001
轮次：20
rb均值: 0.7791043520,rb方差:5.489738941193
rb均值: 3.6126830578,rb方差:7.169084072113
rb均值: -3.4088160992,rb方差:12.016390800476
rb均值: -1.5911220312,rb方差:7.519111633301
rb均值: -0.1070324704,rb方差:11.800100326538
rb均值: -1.9028270245,rb方差:7.821904659271
rb均值: -1.5658330917,rb方差:13.208989143372
时间: 19944.724609 ms
train Classification result: 95.96% ok (used 49984 images)
时间: 1555.826050 ms
Test Classification result: 79.41% ok (used 9984 images)
learn rate：0.0001
轮次：21
rb均值: 0.7934284210,rb方差:5.575552940369
rb均值: 3.6629397869,rb方差:7.263528347015
rb均值: -3.4267106056,rb方差:12.232525825500
rb均值: -1.6193666458,rb方差:7.646870136261
rb均值: -0.1641797721,rb方差:11.971164703369
rb均值: -1.9325962067,rb方差:7.885268211365
rb均值: -1.6605972052,rb方差:13.265716552734
时间: 19883.275391 ms
train Classification result: 96.36% ok (used 49984 images)
时间: 1566.593994 ms
Test Classification result: 79.64% ok (used 9984 images)
learn rate：0.0001
轮次：22
rb均值: 0.7886023521,rb方差:5.626035690308
rb均值: 3.6691143513,rb方差:7.316809654236
rb均值: -3.4639978409,rb方差:12.244820594788
rb均值: -1.6476231813,rb方差:7.712459087372
rb均值: -0.1871936470,rb方差:11.964520454407
rb均值: -1.9374449253,rb方差:7.939944744110
rb均值: -1.6499752998,rb方差:13.451257705688
时间: 20004.205078 ms
train Classification result: 96.61% ok (used 49984 images)
时间: 1563.313965 ms
Test Classification result:81.08% ok (used 9984 images)
learn rate：1e-05
轮次：23
rb均值: 0.7823414207,rb方差:5.631298542023
rb均值: 3.6779696941,rb方差:7.326291084290
rb均值: -3.4713153839,rb方差:12.260807991028
rb均值: -1.6509445906,rb方差:7.716399669647
rb均值: -0.1805491298,rb方差:11.876937866211
rb均值: -1.9555829763,rb方差:8.138381004333
rb均值: -1.6627876759,rb方差:13.509260177612
时间: 19865.951172 ms
train Classification result: 96.84% ok (used 49984 images)
时间: 1561.590942 ms
Test Classification result: 80.91% ok (used 9984 images)
learn rate：1e-05
轮次：24
rb均值: 0.7778691053,rb方差:5.637853145599
rb均值: 3.6839103699,rb方差:7.343599796295
rb均值: -3.4758472443,rb方差:12.273109436035
rb均值: -1.6513724327,rb方差:7.731798648834
rb均值: -0.1821800023,rb方差:11.828219413757
rb均值: -1.9653997421,rb方差:8.259560585022
rb均值: -1.6750459671,rb方差:13.537387847900
时间: 19829.431641 ms
train Classification result: 96.80% ok (used 49984 images)
时间: 1566.438965 ms
Test Classification result: 75.95% ok (used 9984 images)
learn rate：1e-05
轮次：25
rb均值: 0.7756170630,rb方差:5.644399166107
rb均值: 3.6718242168,rb方差:7.348351478577
rb均值: -3.4813792706,rb方差:12.292818069458
rb均值: -1.6526100636,rb方差:7.740932464600
rb均值: -0.1755176932,rb方差:11.933294296265
rb均值: -1.9524576664,rb方差:7.674403190613
rb均值: -1.6790263653,rb方差:13.371887207031
时间: 19854.267578 ms
train Classification result: 96.76% ok (used 49984 images)
时间: 1576.203979 ms
Test Classification result: 79.81% ok (used 9984 images)
learn rate：1e-05
轮次：26
rb均值: 0.7735892534,rb方差:5.652861118317
rb均值: 3.6863250732,rb方差:7.357443809509
rb均值: -3.4883518219,rb方差:12.321460723877
rb均值: -1.6520073414,rb方差:7.762310504913
rb均值: -0.1780894995,rb方差:11.827589035034
rb均值: -1.9692174196,rb方差:8.040866851807
rb均值: -1.6894314289,rb方差:13.462192535400
时间: 20047.675781 ms
train Classification result: 96.98% ok (used 49984 images)
时间: 1562.890991 ms
Test Classification result: 80.70% ok (used 9984 images)
learn rate：1e-05
轮次：27
rb均值: 0.7730028629,rb方差:5.663886070251
rb均值: 3.6962339878,rb方差:7.372319698334
rb均值: -3.4925782681,rb方差:12.335347175598
rb均值: -1.6503198147,rb方差:7.763920307159
rb均值: -0.1908431500,rb方差:11.881231307983
rb均值: -1.9668805599,rb方差:8.041777610779
rb均值: -1.6990214586,rb方差:13.458509445190
时间: 20039.363281 ms
train Classification result: 97.10% ok (used 49984 images)
时间: 1576.229980 ms
Test Classification result: 80.85% ok (used 9984 images)
learn rate：1e-05
轮次：28
rb均值: 0.7713404894,rb方差:5.671066761017
rb均值: 3.7004432678,rb方差:7.388862133026
rb均值: -3.4987366199,rb方差:12.356130599976
rb均值: -1.6501939297,rb方差:7.781978607178
rb均值: -0.1902780682,rb方差:11.800567626953
rb均值: -1.9773398638,rb方差:8.210789680481
rb均值: -1.7096889019,rb方差:13.508257865906
时间: 19995.990234 ms
train Classification result: 97.04% ok (used 49984 images)
时间: 1560.785034 ms
Test Classification result: 81.18% ok (used 9984 images)
使用两个最大池化，也一样：
就是时间上比darknet慢4秒一轮！
