论文总字数:1830字
扩充训练集
使用Cook-Torrance模型生成如下图片
图1-1 Cook-Torrance模型生成图片
扩充训练集识别效果
表2-1 Cook-Torrance模型生成图片训练记录
迭代次数 | Test acc | Test loss | Train loss |
0 | 0.0025 | 6.25586 | 7.02153 |
200 | 0.456875 | 3.10111 | 3.44924 |
400 | 0.7925 | 0.960291 | 3.04043 |
600 | 0.89 | 0.504727 | 0.941217 |
800 | 0.946875 | 0.300117 | 0.619631 |
1000 | 0.976875 | 0.190436 | 0.0840416 |
1200 | 0.966875 | 0.186759 | 0.910046 |
1400 | 0.978125 | 0.130751 | 0.00952794 |
1600 | 0.98125 | 0.10975 | 0.0280898 |
1800 | 0.98375 | 0.164874 | 0.547548 |
2000 | 0.984375 | 0.118263 | 0.00247999 |
2200 | 0.9775 | 0.153099 | 0.00593862 |
2400 | 0.98375 | 0.182122 | 3.10298e-005 |
2600 | 0.9825 | 0.133987 | 0.00013727 |
2800 | 0.9875 | 0.0879754 | 0.0296781 |
Vgg-16在MUCT数据库上的微调
表3-1 MUCT数据库训练记录
迭代次数 | Test acc | Test loss | Train loss |
0 | 0.00125 | 6.09871 | 6.77739 |
40 | 0.03625 | 5.44928 | 5.53218 |
80 | 0.125 | 5.19996 | 5.18764 |
120 | 0.27875 | 4.46124 | 4.29199 |
160 | 0.5825 | 2.79466 | 4.98128 |
200 | 0.865 | 1.10783 | 1.6463 |
240 | 0.97375 | 0.400018 | 0.24024 |
280 | 0.99125 | 0.139569 | 0.315475 |
320 | 0.99625 | 0.0615548 | 0.249584 |
360 | 0.995 | 0.0517763 | 0.116676 |
400 | 0.99375 | 0.0387854 | 0.0700128 |
440 | 0.99875 | 0.0161018 | 0.040963 |
480 | 0.99625 | 0.028108 | 0.0528883 |
520 | 0.99625 | 0.0211374 | 0.0877378 |
560 | 0.995 | 0.0190692 | 0.00144178 |
图3-1 Test accuracy
图3-2 Test loss
图3-2 Train loss
Vgg-16在深度图上的微调
表4-1 原始深度图训练记录
迭代次数 | Test acc | Test loss | Train loss | |||
0 | 0.00625 | 6.82352 | 8.46296 | |||
400 | 0.045833 | 4.95829 | 4.54335 | |||
800 | 0.38125 | 2.87343 | 1.97032 | |||
1200 | 0.529167 | 2.15675 | 0.549867 | |||
1600 | 0.625 | 1.74995 | 0.352125 | |||
2000 | 0.739583 | 1.25592 | 0.0956699 | |||
2400 | 0.710417 | 1.4166 | 0.279212 | |||
2800 | 0.779167 | 1.14456 | 0.00110553 | |||
3200 | 0.81875 | 0.90242 | 0.00173616 | |||
3600 | 0.839583 | 0.837011 | 9.35951e-005 | |||
4000 | 0.845833 | 0.846189 | 0.000595685 | |||
4400 | 0.839583 | 0.824112 | 0.0570602 | |||
4800 | 0.833333 | 0.853865 | 0.000217681 | |||
5200 | 0.845833 | 0.880347 | 0.00291841 | |||
5600 | 0.847917 | 0.810401 | 0.000180236 |
图4-1 Test accuracy
图4-2 Test loss
图4-3 Train loss
剩余内容已隐藏,请支付后下载全文,论文总字数:1830字
该课题毕业论文、开题报告、外文翻译、程序设计、图纸设计等资料可联系客服协助查找;