内容介绍
内容介绍
MAX78000:是一个小型模块 — 66 × 23 mm — 具有 MAX78000、微型 VGA 摄像头、数字麦克风、立体声音频 I/O、microSD 卡插槽、1 MB QSPI RAM、SWD 调试器/编程器USB 端口和锂聚合物电池充电器。还有两个用户 RGB LED 和两个用户按钮以及与 Adafruit Feather 外形兼容的扩展连接器。一个单独的 JTAG 连接器可用于对 RISC-V 内核进行编程和调试。
开发板入门：收到板子后，板子出厂带了个语音识别的DEMO。可识别英文单词“Zero”到“Nine”、“Go”、“Stop”、“Left”、“Right”、“Up”、“Down”、“On”和“Off” . 当它检测到“Go”这个词时，演示会进入数字识别模式，在这种模式下，它会按照扬声器的指令闪烁 LED 灯的次数。也就是说，说“六”后，LED 会闪烁六次。“停止”返回正常模式。串口能显示相关的信息。
安装工具。厂家提供工具是基于Eclipse的工具链，在线安装一共有4G多，很多服务器在国外，需要保持好的网络。包括MinGW、用于 ARM 和 RISC-V 处理器的 GCC 工具链、OpenOCD等。看厂家说明，还可以使用VSCODE作为开发工具。打开安装目录，可以看见官方提供的例程。例程还是蛮丰富的，从入门的hello到外设的例程，还有CNN机器学习的例程。打开官方例程，需要修改例程的两个地方才能正常运行。
1、例程中的Makefile中开发板的选择要由“BOARD ?= EvKit_V1”修改为“BOARD ?= FTHR_RevA”。手头这个板子型号就是FTHR_RevA。
2、修个C/C++Build中的Build command的设置
搭建机器学习的环境。参考https://github.com/MaximIntegratedAI中的文档，使用Anconda创建一个新的python环境（留意环境python要选择Python 3.8.2），安装需要的功能包，注意版本号。
numpy>=1.22,<1.23
PyYAML>=5.1.1
scipy>=1.3.0
librosa>=0.7.2
Pillow>=7
shap>=0.34.0
tk>=0.1.0
torch==1.8.1
torchaudio==0.8.1
torchvision==0.9.1
tensorboard>=2.9.0,<2.10.0
protobuf>=3.20.1,<4.0
numba<0.50.0
opencv-python>=4.4.0
pytsmod>=0.3.3
h5py>=3.7.0参考网络上的大神的文章《美信Maxim78000Evaluation Kit AI开发环境》在WIN10下搭建了python环境，用ai8x-training尝试做一次训练。
入门到放弃。拿到板子，开始的想法是想用板子上的摄像头做一个垃圾分类的项目。通过板子上的摄像头获取垃圾物品的图片，通过机器学习，识别图片进行垃圾分类，然后控制Gpio，模拟控制垃圾桶的开关，实现垃圾分类。这个项目网上挺多资源的，从网上下载了几个G的垃圾图片，用来做训练的资源。然后发现了两个问题：始终没搞懂官方例程中的人脸识别程序的结构，获取摄像头数据后，如何使用机器学习模型？如何使用图片做分类的。更致命的问题是，没有GPU，使用CPU训练速度太慢了，就官方的例程做训练，整个训练过程超过7小时。这样给自己试错的机会就太少了。
2022-11-28 17:32:43,528 - Log file for this run: E:\MAX78000\ai8x-training\logs\2022.11.28-173243\2022.11.28-173243.log
2022-11-28 17:32:43,567 - Optimizer Type: <class 'torch.optim.sgd.SGD'>
2022-11-28 17:32:43,568 - Optimizer Args: {'lr': 0.1, 'momentum': 0.9, 'dampening': 0, 'weight_decay': 0.0001, 'nesterov': False}
2022-11-28 17:32:44,159 - Dataset sizes:
training=54000
validation=6000
test=10000
2022-11-28 17:32:44,160 - Reading compression schedule from: policies/schedule.yaml
2022-11-28 17:32:44,163 -
2022-11-28 17:32:44,164 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-28 17:32:54,596 - Epoch: [0][ 10/ 211] Overall Loss 2.298564 Objective Loss 2.298564 LR 0.100000 Time 1.043111
2022-11-28 17:32:58,981 - Epoch: [0][ 20/ 211] Overall Loss 2.265719 Objective Loss 2.265719 LR 0.100000 Time 0.739323
2022-11-28 17:33:03,232 - Epoch: [0][ 30/ 211] Overall Loss 2.199830 Objective Loss 2.199830 LR 0.100000 Time 0.634570
2022-11-28 17:33:07,517 - Epoch: [0][ 40/ 211] Overall Loss 2.076009 Objective Loss 2.076009 LR 0.100000 Time 0.582991
2022-11-28 17:33:11,869 - Epoch: [0][ 50/ 211] Overall Loss 1.934813 Objective Loss 1.934813 LR 0.100000 Time 0.553440
2022-11-28 17:33:16,155 - Epoch: [0][ 60/ 211] Overall Loss 1.809629 Objective Loss 1.809629 LR 0.100000 Time 0.532626
2022-11-28 17:33:20,413 - Epoch: [0][ 70/ 211] Overall Loss 1.684102 Objective Loss 1.684102 LR 0.100000 Time 0.517360
2022-11-28 17:33:24,715 - Epoch: [0][ 80/ 211] Overall Loss 1.568398 Objective Loss 1.568398 LR 0.100000 Time 0.506446
2022-11-28 17:33:28,997 - Epoch: [0][ 90/ 211] Overall Loss 1.472792 Objective Loss 1.472792 LR 0.100000 Time 0.497747
2022-11-28 17:33:33,556 - Epoch: [0][ 100/ 211] Overall Loss 1.384406 Objective Loss 1.384406 LR 0.100000 Time 0.493550
2022-11-28 17:33:38,080 - Epoch: [0][ 110/ 211] Overall Loss 1.313016 Objective Loss 1.313016 LR 0.100000 Time 0.489809
2022-11-28 17:33:42,663 - Epoch: [0][ 120/ 211] Overall Loss 1.250540 Objective Loss 1.250540 LR 0.100000 Time 0.487172
2022-11-28 17:33:46,950 - Epoch: [0][ 130/ 211] Overall Loss 1.191245 Objective Loss 1.191245 LR 0.100000 Time 0.482671
2022-11-28 17:33:51,245 - Epoch: [0][ 140/ 211] Overall Loss 1.133978 Objective Loss 1.133978 LR 0.100000 Time 0.478862
2022-11-28 17:33:55,524 - Epoch: [0][ 150/ 211] Overall Loss 1.083477 Objective Loss 1.083477 LR 0.100000 Time 0.475469
2022-11-28 17:33:59,784 - Epoch: [0][ 160/ 211] Overall Loss 1.038986 Objective Loss 1.038986 LR 0.100000 Time 0.472368
2022-11-28 17:34:04,060 - Epoch: [0][ 170/ 211] Overall Loss 0.998710 Objective Loss 0.998710 LR 0.100000 Time 0.469732
2022-11-28 17:34:08,315 - Epoch: [0][ 180/ 211] Overall Loss 0.961280 Objective Loss 0.961280 LR 0.100000 Time 0.467267
2022-11-28 17:34:12,609 - Epoch: [0][ 190/ 211] Overall Loss 0.927787 Objective Loss 0.927787 LR 0.100000 Time 0.465272
2022-11-28 17:34:16,875 - Epoch: [0][ 200/ 211] Overall Loss 0.896116 Objective Loss 0.896116 LR 0.100000 Time 0.463331
2022-11-28 17:34:21,197 - Epoch: [0][ 210/ 211] Overall Loss 0.866394 Objective Loss 0.866394 Top1 92.187500 Top5 100.000000 LR 0.100000 Time 0.461708
2022-11-28 17:34:21,598 - Epoch: [0][ 211/ 211] Overall Loss 0.863859 Objective Loss 0.863859 Top1 90.927419 Top5 99.798387 LR 0.100000 Time 0.461420
2022-11-28 17:34:22,150 - --- validate (epoch=0)-----------
2022-11-28 17:34:22,150 - 6000 samples (256 per mini-batch)
2022-11-28 17:34:29,647 - Epoch: [0][ 10/ 24] Loss 0.283329 Top1 91.328125 Top5 99.726562
2022-11-28 17:34:31,309 - Epoch: [0][ 20/ 24] Loss 0.281354 Top1 91.406250 Top5 99.687500
2022-11-28 17:34:31,875 - Epoch: [0][ 24/ 24] Loss 0.282949 Top1 91.433333 Top5 99.716667
2022-11-28 17:34:32,415 - ==> Top1: 91.433 Top5: 99.717 Loss: 0.283
2022-11-28 17:34:32,416 - ==> Confusion:
[[564 0 4 2 1 1 8 1 21 3]
[ 0 665 6 3 2 0 7 3 2 0]
[ 1 0 525 15 5 1 4 9 24 2]
[ 0 1 10 557 0 3 0 2 9 1]
[ 3 2 5 2 503 1 10 8 9 22]
[ 1 2 2 15 1 438 17 3 30 9]
[ 0 1 3 0 5 1 602 0 18 1]
[ 0 4 21 29 2 2 0 547 4 16]
[ 1 1 4 3 3 2 13 1 550 6]
[ 2 3 9 12 15 6 3 7 23 535]]
2022-11-28 17:34:32,422 - ==> Best [Top1: 91.433 Top5: 99.717 Sparsity:0.00 Params: 71148 on epoch: 0]
2022-11-28 17:34:32,422 - Saving checkpoint to: logs\2022.11.28-173243\checkpoint.pth.tar
2022-11-28 17:34:32,483 -
2022-11-28 17:34:32,483 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-28 17:34:42,455 - Epoch: [1][ 10/ 211] Overall Loss 0.303115 Objective Loss 0.303115 LR 0.100000 Time 0.996934
2022-11-28 17:34:46,806 - Epoch: [1][ 20/ 211] Overall Loss 0.292789 Objective Loss 0.292789 LR 0.100000 Time 0.715986
2022-11-28 17:34:51,059 - Epoch: [1][ 30/ 211] Overall Loss 0.288909 Objective Loss 0.288909 LR 0.100000 Time 0.619078
2022-11-28 17:34:55,471 - Epoch: [1][ 40/ 211] Overall Loss 0.286541 Objective Loss 0.286541 LR 0.100000 Time 0.574614
2022-11-28 17:34:59,707 - Epoch: [1][ 50/ 211] Overall Loss 0.290066 Objective Loss 0.290066 LR 0.100000 Time 0.544404
2022-11-28 17:35:03,996 - Epoch: [1][ 60/ 211] Overall Loss 0.282046 Objective Loss 0.282046 LR 0.100000 Time 0.525162
2022-11-28 17:35:08,304 - Epoch: [1][ 70/ 211] Overall Loss 0.272592 Objective Loss 0.272592 LR 0.100000 Time 0.511675
2022-11-28 17:35:12,609 - Epoch: [1][ 80/ 211] Overall Loss 0.265767 Objective Loss 0.265767 LR 0.100000 Time 0.501509
2022-11-28 17:35:16,864 - Epoch: [1][ 90/ 211] Overall Loss 0.259592 Objective Loss 0.259592 LR 0.100000 Time 0.493048
2022-11-28 17:35:21,260 - Epoch: [1][ 100/ 211] Overall Loss 0.254881 Objective Loss 0.254881 LR 0.100000 Time 0.487706
2022-11-28 17:35:25,552 - Epoch: [1][ 110/ 211] Overall Loss 0.253164 Objective Loss 0.253164 LR 0.100000 Time 0.482383
2022-11-28 17:35:29,769 - Epoch: [1][ 120/ 211] Overall Loss 0.249848 Objective Loss 0.249848 LR 0.100000 Time 0.477332
2022-11-28 17:35:33,950 - Epoch: [1][ 130/ 211] Overall Loss 0.246969 Objective Loss 0.246969 LR 0.100000 Time 0.472767
2022-11-28 17:35:38,209 - Epoch: [1][ 140/ 211] Overall Loss 0.243982 Objective Loss 0.243982 LR 0.100000 Time 0.469416
2022-11-28 17:35:42,414 - Epoch: [1][ 150/ 211] Overall Loss 0.240281 Objective Loss 0.240281 LR 0.100000 Time 0.466154
2022-11-28 17:35:46,660 - Epoch: [1][ 160/ 211] Overall Loss 0.237530 Objective Loss 0.237530 LR 0.100000 Time 0.463554
2022-11-28 17:35:50,871 - Epoch: [1][ 170/ 211] Overall Loss 0.233607 Objective Loss 0.233607 LR 0.100000 Time 0.461055
2022-11-28 17:35:55,097 - Epoch: [1][ 180/ 211] Overall Loss 0.231156 Objective Loss 0.231156 LR 0.100000 Time 0.458912
2022-11-28 17:35:59,281 - Epoch: [1][ 190/ 211] Overall Loss 0.229752 Objective Loss 0.229752 LR 0.100000 Time 0.456779
2022-11-28 17:36:03,531 - Epoch: [1][ 200/ 211] Overall Loss 0.228390 Objective Loss 0.228390 LR 0.100000 Time 0.455178
2022-11-28 17:36:07,763 - Epoch: [1][ 210/ 211] Overall Loss 0.226308 Objective Loss 0.226308 Top1 94.140625 Top5 100.000000 LR 0.100000 Time 0.453649
2022-11-28 17:36:08,165 - Epoch: [1][ 211/ 211] Overall Loss 0.226081 Objective Loss 0.226081 Top1 94.758065 Top5 99.395161 LR 0.100000 Time 0.453399
2022-11-28 17:36:08,708 - --- validate (epoch=1)-----------
2022-11-28 17:36:08,708 - 6000 samples (256 per mini-batch)
2022-11-28 17:36:15,978 - Epoch: [1][ 10/ 24] Loss 0.169041 Top1 94.960938 Top5 99.765625
2022-11-28 17:38:31,130 - Epoch: [1][ 20/ 24] Loss 0.173639 Top1 94.765625 Top5 99.785156
2022-11-28 17:38:31,683 - Epoch: [1][ 24/ 24] Loss 0.172894 Top1 94.750000 Top5 99.783333
2022-11-28 17:38:32,225 - ==> Top1: 94.750 Top5: 99.783 Loss: 0.173
2022-11-28 17:38:32,226 - ==> Confusion:
[[595 0 1 0 0 0 4 0 5 0]
[ 1 666 8 1 4 0 2 5 0 1]
[ 5 0 546 5 2 1 1 9 14 3]
[ 3 0 8 556 0 1 1 3 5 6]
[ 2 0 4 1 528 0 5 2 3 20]
[ 6 0 5 7 1 462 6 2 19 10]
[ 6 0 0 0 2 1 610 0 12 0]
[ 0 1 9 6 3 0 0 588 3 15]
[ 4 0 0 0 5 0 9 1 561 4]
[ 5 0 0 7 6 4 1 7 12 573]]
2022-11-28 17:38:32,228 - ==> Best [Top1: 94.750 Top5: 99.783 Sparsity:0.00 Params: 71148 on epoch: 1]
2022-11-28 17:38:32,229 - Saving checkpoint to: logs\2022.11.28-173243\checkpoint.pth.tar
2022-11-28 17:38:32,239 -
2022-11-28 17:38:32,240 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-28 17:38:42,156 - Epoch: [2][ 10/ 211] Overall Loss 0.162670 Objective Loss 0.162670 LR 0.100000 Time 0.991549
2022-11-28 17:38:46,376 - Epoch: [2][ 20/ 211] Overall Loss 0.163434 Objective Loss 0.163434 LR 0.100000 Time 0.706760
2022-11-28 17:38:50,723 - Epoch: [2][ 30/ 211] Overall Loss 0.169471 Objective Loss 0.169471 LR 0.100000 Time 0.616070
2022-11-28 17:38:54,955 - Epoch: [2][ 40/ 211] Overall Loss 0.171583 Objective Loss 0.171583 LR 0.100000 Time 0.567832
2022-11-28 17:38:59,168 - Epoch: [2][ 50/ 211] Overall Loss 0.175466 Objective Loss 0.175466 LR 0.100000 Time 0.538540
2022-11-28 17:39:03,386 - Epoch: [2][ 60/ 211] Overall Loss 0.177056 Objective Loss 0.177056 LR 0.100000 Time 0.519062
2022-11-28 17:39:07,682 - Epoch: [2][ 70/ 211] Overall Loss 0.175469 Objective Loss 0.175469 LR 0.100000 Time 0.506275
2022-11-28 17:39:11,976 - Epoch: [2][ 80/ 211] Overall Loss 0.172752 Objective Loss 0.172752 LR 0.100000 Time 0.496648
2022-11-28 17:39:16,241 - Epoch: [2][ 90/ 211] Overall Loss 0.171963 Objective Loss 0.171963 LR 0.100000 Time 0.488838
2022-11-28 17:39:20,490 - Epoch: [2][ 100/ 211] Overall Loss 0.171959 Objective Loss 0.171959 LR 0.100000 Time 0.482450
2022-11-28 17:39:24,739 - Epoch: [2][ 110/ 211] Overall Loss 0.171803 Objective Loss 0.171803 LR 0.100000 Time 0.477206
2022-11-28 17:39:28,952 - Epoch: [2][ 120/ 211] Overall Loss 0.171238 Objective Loss 0.171238 LR 0.100000 Time 0.472537
2022-11-28 17:39:33,163 - Epoch: [2][ 130/ 211] Overall Loss 0.171115 Objective Loss 0.171115 LR 0.100000 Time 0.468570
2022-11-28 17:39:37,404 - Epoch: [2][ 140/ 211] Overall Loss 0.171511 Objective Loss 0.171511 LR 0.100000 Time 0.465384
2022-11-28 17:39:41,645 - Epoch: [2][ 150/ 211] Overall Loss 0.170378 Objective Loss 0.170378 LR 0.100000 Time 0.462623
2022-11-28 17:39:45,879 - Epoch: [2][ 160/ 211] Overall Loss 0.169934 Objective Loss 0.169934 LR 0.100000 Time 0.460170
2022-11-28 17:39:50,078 - Epoch: [2][ 170/ 211] Overall Loss 0.170061 Objective Loss 0.170061 LR 0.100000 Time 0.457800
2022-11-28 17:39:54,327 - Epoch: [2][ 180/ 211] Overall Loss 0.168993 Objective Loss 0.168993 LR 0.100000 Time 0.455964
2022-11-28 17:39:58,554 - Epoch: [2][ 190/ 211] Overall Loss 0.167059 Objective Loss 0.167059 LR 0.100000 Time 0.454207
2022-11-28 17:40:02,816 - Epoch: [2][ 200/ 211] Overall Loss 0.166115 Objective Loss 0.166115 LR 0.100000 Time 0.452805
2022-11-28 17:40:07,072 - Epoch: [2][ 210/ 211] Overall Loss 0.164306 Objective Loss 0.164306 Top1 96.484375 Top5 100.000000 LR 0.100000 Time 0.451507
2022-11-28 17:40:07,468 - Epoch: [2][ 211/ 211] Overall Loss 0.164035 Objective Loss 0.164035 Top1 96.774194 Top5 99.798387 LR 0.100000 Time 0.451244
2022-11-28 17:40:08,038 - --- validate (epoch=2)-----------
2022-11-28 17:40:08,038 - 6000 samples (256 per mini-batch)
2022-11-28 17:40:15,657 - Epoch: [2][ 10/ 24] Loss 0.131910 Top1 96.445312 Top5 99.804688
2022-11-28 17:40:17,330 - Epoch: [2][ 20/ 24] Loss 0.134751 Top1 96.269531 Top5 99.882812
2022-11-28 17:40:17,890 - Epoch: [2][ 24/ 24] Loss 0.134910 Top1 96.166667 Top5 99.900000
2022-11-28 17:40:18,431 - ==> Top1: 96.167 Top5: 99.900 Loss: 0.135
2022-11-28 17:40:18,431 - ==> Confusion:
[[595 0 3 0 0 1 5 0 1 0]
[ 0 685 2 0 1 0 0 0 0 0]
[ 1 1 560 2 1 4 2 7 6 2]
[ 0 2 7 557 0 8 1 3 5 0]
[ 0 1 4 0 546 1 5 1 0 7]
[ 1 0 3 2 2 503 6 0 1 0]
[ 3 2 0 0 1 3 619 0 3 0]
[ 0 3 10 0 4 3 0 599 2 4]
[ 2 2 1 0 3 8 10 4 549 5]
[ 4 6 1 4 13 12 1 8 9 557]]
2022-11-28 17:40:18,433 - ==> Best [Top1: 96.167 Top5: 99.900 Sparsity:0.00 Params: 71148 on epoch: 2]
2022-11-28 17:40:18,434 - Saving checkpoint to: logs\2022.11.28-173243\checkpoint.pth.tar
2022-11-28 17:40:18,462 -
2022-11-28 17:40:18,462 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-28 17:40:28,648 - Epoch: [3][ 10/ 211] Overall Loss 0.135862 Objective Loss 0.135862 LR 0.100000 Time 1.018377
2022-11-28 17:40:32,899 - Epoch: [3][ 20/ 211] Overall Loss 0.138279 Objective Loss 0.138279 LR 0.100000 Time 0.721720
2022-11-28 17:40:37,147 - Epoch: [3][ 30/ 211] Overall Loss 0.138205 Objective Loss 0.138205 LR 0.100000 Time 0.622735
2022-11-28 17:40:41,643 - Epoch: [3][ 40/ 211] Overall Loss 0.137411 Objective Loss 0.137411 LR 0.100000 Time 0.579476
2022-11-28 17:40:46,032 - Epoch: [3][ 50/ 211] Overall Loss 0.141440 Objective Loss 0.141440 LR 0.100000 Time 0.551346
2022-11-28 17:40:50,770 - Epoch: [3][ 60/ 211] Overall Loss 0.143205 Objective Loss 0.143205 LR 0.100000 Time 0.538427
2022-11-28 17:40:55,172 - Epoch: [3][ 70/ 211] Overall Loss 0.143112 Objective Loss 0.143112 LR 0.100000 Time 0.524377
2022-11-28 17:40:59,533 - Epoch: [3][ 80/ 211] Overall Loss 0.142289 Objective Loss 0.142289 LR 0.100000 Time 0.513334
2022-11-28 17:41:03,833 - Epoch: [3][ 90/ 211] Overall Loss 0.140776 Objective Loss 0.140776 LR 0.100000 Time 0.504058
2022-11-28 17:41:08,179 - Epoch: [3][ 100/ 211] Overall Loss 0.141807 Objective Loss 0.141807 LR 0.100000 Time 0.497106
2022-11-28 17:41:12,483 - Epoch: [3][ 110/ 211] Overall Loss 0.144942 Objective Loss 0.144942 LR 0.100000 Time 0.491028
2022-11-28 17:41:16,751 - Epoch: [3][ 120/ 211] Overall Loss 0.144196 Objective Loss 0.144196 LR 0.100000 Time 0.485681
2022-11-28 17:41:21,042 - Epoch: [3][ 130/ 211] Overall Loss 0.143078 Objective Loss 0.143078 LR 0.100000 Time 0.481317
2022-11-28 17:41:25,327 - Epoch: [3][ 140/ 211] Overall Loss 0.140382 Objective Loss 0.140382 LR 0.100000 Time 0.477548
2022-11-28 17:41:29,718 - Epoch: [3][ 150/ 211] Overall Loss 0.138152 Objective Loss 0.138152 LR 0.100000 Time 0.474987
2022-11-28 17:41:33,977 - Epoch: [3][ 160/ 211] Overall Loss 0.136183 Objective Loss 0.136183 LR 0.100000 Time 0.471904
2022-11-28 17:41:38,333 - Epoch: [3][ 170/ 211] Overall Loss 0.134215 Objective Loss 0.134215 LR 0.100000 Time 0.469770
2022-11-28 17:41:42,621 - Epoch: [3][ 180/ 211] Overall Loss 0.133231 Objective Loss 0.133231 LR 0.100000 Time 0.467486
2022-11-28 17:41:46,838 - Epoch: [3][ 190/ 211] Overall Loss 0.132332 Objective Loss 0.132332 LR 0.100000 Time 0.465064
2022-11-28 17:41:51,067 - Epoch: [3][ 200/ 211] Overall Loss 0.132762 Objective Loss 0.132762 LR 0.100000 Time 0.462960
2022-11-28 17:41:55,306 - Epoch: [3][ 210/ 211] Overall Loss 0.132196 Objective Loss 0.132196 Top1 96.093750 Top5 100.000000 LR 0.100000 Time 0.461093
2022-11-28 17:41:55,702 - Epoch: [3][ 211/ 211] Overall Loss 0.132407 Objective Loss 0.132407 Top1 95.161290 Top5 99.798387 LR 0.100000 Time 0.460785
2022-11-28 17:41:56,254 - --- validate (epoch=3)-----------
2022-11-28 17:41:56,254 - 6000 samples (256 per mini-batch)
2022-11-28 17:42:03,501 - Epoch: [3][ 10/ 24] Loss 0.116774 Top1 96.484375 Top5 99.882812
2022-11-28 17:42:05,133 - Epoch: [3][ 20/ 24] Loss 0.110772 Top1 96.699219 Top5 99.921875
2022-11-28 17:42:05,696 - Epoch: [3][ 24/ 24] Loss 0.112113 Top1 96.583333 Top5 99.933333
2022-11-28 17:42:06,243 - ==> Top1: 96.583 Top5: 99.933 Loss: 0.112
2022-11-28 17:42:06,244 - ==> Confusion:
[[596 0 1 0 1 0 1 0 1 5]
[ 0 682 4 1 0 0 0 1 0 0]
[ 1 2 569 2 1 0 0 7 2 2]
[ 0 0 7 559 2 3 0 6 3 3]
[ 0 1 3 0 547 2 0 3 1 8]
[ 3 1 1 1 2 500 3 0 3 4]
[ 4 5 2 0 4 12 598 0 5 1]
[ 0 3 3 3 3 0 0 611 0 2]
[ 5 2 5 0 5 3 2 3 552 7]
[ 2 3 0 3 13 3 0 9 1 581]]
2022-11-28 17:42:06,246 - ==> Best [Top1: 96.583 Top5: 99.933 Sparsity:0.00 Params: 71148 on epoch: 3]
2022-11-28 17:42:06,246 - Saving checkpoint to: logs\2022.11.28-173243\checkpoint.pth.tar
2022-11-28 17:42:06,251 -
2022-11-28 17:42:06,251 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-28 17:42:16,146 - Epoch: [4][ 10/ 211] Overall Loss 0.105769 Objective Loss 0.105769 LR 0.100000 Time 0.989454
2022-11-28 17:42:20,441 - Epoch: [4][ 20/ 211] Overall Loss 0.112949 Objective Loss 0.112949 LR 0.100000 Time 0.709453
2022-11-28 17:42:24,725 - Epoch: [4][ 30/ 211] Overall Loss 0.117931 Objective Loss 0.117931 LR 0.100000 Time 0.615787
2022-11-28 17:42:28,958 - Epoch: [4][ 40/ 211] Overall Loss 0.116802 Objective Loss 0.116802 LR 0.100000 Time 0.567657
2022-11-28 17:42:33,220 - Epoch: [4][ 50/ 211] Overall Loss 0.114094 Objective Loss 0.114094 LR 0.100000 Time 0.539358
2022-11-28 17:42:37,430 - Epoch: [4][ 60/ 211] Overall Loss 0.111658 Objective Loss 0.111658 LR 0.100000 Time 0.519627
2022-11-28 17:42:41,749 - Epoch: [4][ 70/ 211] Overall Loss 0.114227 Objective Loss 0.114227 LR 0.100000 Time 0.507087
2022-11-28 17:42:46,050 - Epoch: [4][ 80/ 211] Overall Loss 0.115709 Objective Loss 0.115709 LR 0.100000 Time 0.497457
2022-11-28 17:42:50,377 - Epoch: [4][ 90/ 211] Overall Loss 0.114733 Objective Loss 0.114733 LR 0.100000 Time 0.490234
2022-11-28 17:42:54,668 - Epoch: [4][ 100/ 211] Overall Loss 0.114030 Objective Loss 0.114030 LR 0.100000 Time 0.484116
2022-11-28 17:42:59,000 - Epoch: [4][ 110/ 211] Overall Loss 0.114818 Objective Loss 0.114818 LR 0.100000 Time 0.479473
2022-11-28 17:43:03,287 - Epoch: [4][ 120/ 211] Overall Loss 0.115313 Objective Loss 0.115313 LR 0.100000 Time 0.475229
2022-11-28 17:43:07,533 - Epoch: [4][ 130/ 211] Overall Loss 0.115641 Objective Loss 0.115641 LR 0.100000 Time 0.471332
2022-11-28 17:43:11,841 - Epoch: [4][ 140/ 211] Overall Loss 0.117132 Objective Loss 0.117132 LR 0.100000 Time 0.468440
2022-11-28 17:43:16,106 - Epoch: [4][ 150/ 211] Overall Loss 0.116808 Objective Loss 0.116808 LR 0.100000 Time 0.465635
2022-11-28 17:43:20,435 - Epoch: [4][ 160/ 211] Overall Loss 0.116517 Objective Loss 0.116517 LR 0.100000 Time 0.463585
2022-11-28 17:43:24,688 - Epoch: [4][ 170/ 211] Overall Loss 0.116621 Objective Loss 0.116621 LR 0.100000 Time 0.461319
2022-11-28 17:43:28,973 - Epoch: [4][ 180/ 211] Overall Loss 0.115962 Objective Loss 0.115962 LR 0.100000 Time 0.459483
2022-11-28 17:43:33,223 - Epoch: [4][ 190/ 211] Overall Loss 0.116119 Objective Loss 0.116119 LR 0.100000 Time 0.457660
2022-11-28 17:43:37,444 - Epoch: [4][ 200/ 211] Overall Loss 0.116018 Objective Loss 0.116018 LR 0.100000 Time 0.455881
2022-11-28 17:43:41,738 - Epoch: [4][ 210/ 211] Overall Loss 0.115201 Objective Loss 0.115201 Top1 96.875000 Top5 100.000000 LR 0.100000 Time 0.454622
2022-11-28 17:43:42,138 - Epoch: [4][ 211/ 211] Overall Loss 0.115169 Objective Loss 0.115169 Top1 96.572581 Top5 100.000000 LR 0.100000 Time 0.454363
2022-11-28 17:43:42,682 - --- validate (epoch=4)-----------
2022-11-28 17:43:42,683 - 6000 samples (256 per mini-batch)
2022-11-28 17:43:49,923 - Epoch: [4][ 10/ 24] Loss 0.092024 Top1 96.875000 Top5 99.960938
2022-11-28 17:43:51,585 - Epoch: [4][ 20/ 24] Loss 0.092981 Top1 96.953125 Top5 99.980469
2022-11-28 17:43:52,135 - Epoch: [4][ 24/ 24] Loss 0.095692 Top1 96.933333 Top5 99.966667
2022-11-28 17:43:52,670 - ==> Top1: 96.933 Top5: 99.967 Loss: 0.096
2022-11-28 17:43:52,670 - ==> Confusion:
[[592 0 3 1 0 2 4 0 3 0]
[ 0 681 5 0 0 0 1 1 0 0]
[ 0 2 567 1 0 0 1 11 4 0]
[ 1 0 5 563 0 3 0 8 3 0]
[ 0 1 2 0 535 0 2 7 0 18]
[ 2 1 0 1 0 500 6 0 5 3]
[ 8 1 1 0 1 2 616 0 2 0]
[ 0 4 7 0 1 0 0 610 0 3]
[ 1 1 5 1 1 1 6 1 562 5]
[ 1 1 1 0 6 4 2 6 4 590]]
2022-11-28 17:43:52,672 - ==> Best [Top1: 96.933 Top5: 99.967 Sparsity:0.00 Params: 71148 on epoch: 4]
2022-11-28 17:43:52,673 - Saving checkpoint to: logs\2022.11.28-173243\checkpoint.pth.tar
2022-11-28 17:43:52,677 -
2022-11-28 17:43:52,677 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-28 17:44:02,563 - Epoch: [5][ 10/ 211] Overall Loss 0.105496 Objective Loss 0.105496 LR 0.100000 Time 0.988457
2022-11-28 17:44:06,888 - Epoch: [5][ 20/ 211] Overall Loss 0.114271 Objective Loss 0.114271 LR 0.100000 Time 0.710450
2022-11-28 17:44:11,201 - Epoch: [5][ 30/ 211] Overall Loss 0.109886 Objective Loss 0.109886 LR 0.100000 Time 0.617383
2022-11-28 17:44:15,425 - Epoch: [5][ 40/ 211] Overall Loss 0.109347 Objective Loss 0.109347 LR 0.100000 Time 0.568654
2022-11-28 17:44:19,705 - Epoch: [5][ 50/ 211] Overall Loss 0.105478 Objective Loss 0.105478 LR 0.100000 Time 0.540515
2022-11-28 17:44:23,989 - Epoch: [5][ 60/ 211] Overall Loss 0.109943 Objective Loss 0.109943 LR 0.100000 Time 0.521838
2022-11-28 17:44:28,274 - Epoch: [5][ 70/ 211] Overall Loss 0.115873 Objective Loss 0.115873 LR 0.100000 Time 0.508483
2022-11-28 17:44:32,505 - Epoch: [5][ 80/ 211] Overall Loss 0.115535 Objective Loss 0.115535 LR 0.100000 Time 0.497806
2022-11-28 17:44:36,738 - Epoch: [5][ 90/ 211] Overall Loss 0.116393 Objective Loss 0.116393 LR 0.100000 Time 0.489524
2022-11-28 17:44:41,033 - Epoch: [5][ 100/ 211] Overall Loss 0.117968 Objective Loss 0.117968 LR 0.100000 Time 0.483507
2022-11-28 17:44:45,275 - Epoch: [5][ 110/ 211] Overall Loss 0.117010 Objective Loss 0.117010 LR 0.100000 Time 0.478103
2022-11-28 17:44:49,535 - Epoch: [5][ 120/ 211] Overall Loss 0.116879 Objective Loss 0.116879 LR 0.100000 Time 0.473750
2022-11-28 17:44:53,861 - Epoch: [5][ 130/ 211] Overall Loss 0.116730 Objective Loss 0.116730 LR 0.100000 Time 0.470580
2022-11-28 17:44:58,178 - Epoch: [5][ 140/ 211] Overall Loss 0.114929 Objective Loss 0.114929 LR 0.100000 Time 0.467799
2022-11-28 17:45:02,497 - Epoch: [5][ 150/ 211] Overall Loss 0.115214 Objective Loss 0.115214 LR 0.100000 Time 0.465402
2022-11-28 17:45:06,807 - Epoch: [5][ 160/ 211] Overall Loss 0.114155 Objective Loss 0.114155 LR 0.100000 Time 0.463249
2022-11-28 17:45:11,111 - Epoch: [5][ 170/ 211] Overall Loss 0.114295 Objective Loss 0.114295 LR 0.100000 Time 0.461308
2022-11-28 17:45:15,388 - Epoch: [5][ 180/ 211] Overall Loss 0.114194 Objective Loss 0.114194 LR 0.100000 Time 0.459427
2022-11-28 17:45:19,597 - Epoch: [5][ 190/ 211] Overall Loss 0.112834 Objective Loss 0.112834 LR 0.100000 Time 0.457393
2022-11-28 17:45:23,936 - Epoch: [5][ 200/ 211] Overall Loss 0.112101 Objective Loss 0.112101 LR 0.100000 Time 0.456220
2022-11-28 17:45:28,217 - Epoch: [5][ 210/ 211] Overall Loss 0.111618 Objective Loss 0.111618 Top1 95.312500 Top5 100.000000 LR 0.100000 Time 0.454869
2022-11-28 17:45:28,627 - Epoch: [5][ 211/ 211] Overall Loss 0.111687 Objective Loss 0.111687 Top1 95.564516 Top5 100.000000 LR 0.100000 Time 0.454652
2022-11-28 17:45:29,185 - --- validate (epoch=5)-----------
2022-11-28 17:45:29,186 - 6000 samples (256 per mini-batch)
2022-11-28 17:45:36,388 - Epoch: [5][ 10/ 24] Loss 0.102077 Top1 97.070312 Top5 99.921875
2022-11-28 17:45:38,022 - Epoch: [5][ 20/ 24] Loss 0.105979 Top1 96.621094 Top5 99.960938
2022-11-28 17:45:38,589 - Epoch: [5][ 24/ 24] Loss 0.101846 Top1 96.750000 Top5 99.966667
2022-11-28 17:45:39,128 - ==> Top1: 96.750 Top5: 99.967 Loss: 0.102
2022-11-28 17:45:39,128 - ==> Confusion:
[[596 0 1 0 0 0 5 0 1 2]
[ 1 678 2 0 1 1 2 3 0 0]
[ 2 0 562 2 2 0 0 3 13 2]
[ 1 0 4 558 0 9 0 1 6 4]
[ 0 0 2 0 544 1 2 1 3 12]
[ 1 0 1 1 0 508 3 0 2 2]
[ 2 1 0 0 2 2 619 0 5 0]
[ 0 5 9 8 8 4 0 578 0 13]
[ 1 0 0 1 1 1 8 0 567 5]
[ 0 1 0 1 10 3 0 1 4 595]]
………………………………
2022-11-29 00:24:37,209 - Epoch: [193][ 150/ 211] Overall Loss 0.040234 Objective Loss 0.040234 LR 0.000100 Time 0.556219
2022-11-29 00:24:42,415 - Epoch: [193][ 160/ 211] Overall Loss 0.040418 Objective Loss 0.040418 LR 0.000100 Time 0.553994
2022-11-29 00:24:47,603 - Epoch: [193][ 170/ 211] Overall Loss 0.041247 Objective Loss 0.041247 LR 0.000100 Time 0.551924
2022-11-29 00:24:52,860 - Epoch: [193][ 180/ 211] Overall Loss 0.040925 Objective Loss 0.040925 LR 0.000100 Time 0.550461
2022-11-29 00:24:58,021 - Epoch: [193][ 190/ 211] Overall Loss 0.041321 Objective Loss 0.041321 LR 0.000100 Time 0.548649
2022-11-29 00:25:03,154 - Epoch: [193][ 200/ 211] Overall Loss 0.041075 Objective Loss 0.041075 LR 0.000100 Time 0.546878
2022-11-29 00:25:08,327 - Epoch: [193][ 210/ 211] Overall Loss 0.040713 Objective Loss 0.040713 Top1 99.218750 Top5 100.000000 LR 0.000100 Time 0.545461
2022-11-29 00:25:08,824 - Epoch: [193][ 211/ 211] Overall Loss 0.040734 Objective Loss 0.040734 Top1 98.991935 Top5 100.000000 LR 0.000100 Time 0.545229
2022-11-29 00:25:09,354 - --- validate (epoch=193)-----------
2022-11-29 00:25:09,355 - 6000 samples (256 per mini-batch)
2022-11-29 00:25:17,403 - Epoch: [193][ 10/ 24] Loss 0.038751 Top1 99.023438 Top5 100.000000
2022-11-29 00:25:20,020 - Epoch: [193][ 20/ 24] Loss 0.042008 Top1 98.828125 Top5 99.980469
2022-11-29 00:25:20,917 - Epoch: [193][ 24/ 24] Loss 0.044995 Top1 98.800000 Top5 99.983333
2022-11-29 00:25:21,448 - ==> Top1: 98.800 Top5: 99.983 Loss: 0.045
2022-11-29 00:25:21,449 - ==> Confusion:
[[599 0 1 0 0 0 2 0 1 2]
[ 0 686 0 0 0 0 0 2 0 0]
[ 0 0 579 1 0 0 1 3 1 1]
[ 0 0 3 576 0 0 0 2 1 1]
[ 0 1 0 0 555 0 1 0 0 8]
[ 0 1 0 2 0 508 5 0 2 0]
[ 1 2 0 0 3 0 625 0 0 0]
[ 0 3 3 1 0 0 0 618 0 0]
[ 0 0 1 0 1 1 2 0 578 1]
[ 1 2 0 0 3 1 0 2 2 604]]
2022-11-29 00:25:21,451 - ==> Best [Top1: 99.067 Top5: 100.000 Sparsity:0.00 Params: 71148 on epoch: 173]
2022-11-29 00:25:21,451 - Saving checkpoint to: logs\2022.11.28-173243\qat_checkpoint.pth.tar
2022-11-29 00:25:21,454 -
2022-11-29 00:25:21,454 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-29 00:25:32,195 - Epoch: [194][ 10/ 211] Overall Loss 0.040329 Objective Loss 0.040329 LR 0.000100 Time 1.074028
2022-11-29 00:25:37,387 - Epoch: [194][ 20/ 211] Overall Loss 0.040542 Objective Loss 0.040542 LR 0.000100 Time 0.796570
2022-11-29 00:25:42,581 - Epoch: [194][ 30/ 211] Overall Loss 0.040781 Objective Loss 0.040781 LR 0.000100 Time 0.704151
2022-11-29 00:25:47,757 - Epoch: [194][ 40/ 211] Overall Loss 0.040673 Objective Loss 0.040673 LR 0.000100 Time 0.657492
2022-11-29 00:25:52,961 - Epoch: [194][ 50/ 211] Overall Loss 0.042351 Objective Loss 0.042351 LR 0.000100 Time 0.630075
2022-11-29 00:25:58,191 - Epoch: [194][ 60/ 211] Overall Loss 0.041670 Objective Loss 0.041670 LR 0.000100 Time 0.612213
2022-11-29 00:26:03,403 - Epoch: [194][ 70/ 211] Overall Loss 0.041011 Objective Loss 0.041011 LR 0.000100 Time 0.599198
2022-11-29 00:26:08,571 - Epoch: [194][ 80/ 211] Overall Loss 0.041417 Objective Loss 0.041417 LR 0.000100 Time 0.588863
2022-11-29 00:26:13,754 - Epoch: [194][ 90/ 211] Overall Loss 0.041389 Objective Loss 0.041389 LR 0.000100 Time 0.581024
2022-11-29 00:26:18,946 - Epoch: [194][ 100/ 211] Overall Loss 0.041455 Objective Loss 0.041455 LR 0.000100 Time 0.574843
2022-11-29 00:26:24,120 - Epoch: [194][ 110/ 211] Overall Loss 0.041739 Objective Loss 0.041739 LR 0.000100 Time 0.569613
2022-11-29 00:26:29,298 - Epoch: [194][ 120/ 211] Overall Loss 0.041621 Objective Loss 0.041621 LR 0.000100 Time 0.565297
2022-11-29 00:26:34,447 - Epoch: [194][ 130/ 211] Overall Loss 0.041431 Objective Loss 0.041431 LR 0.000100 Time 0.561422
2022-11-29 00:26:39,634 - Epoch: [194][ 140/ 211] Overall Loss 0.041545 Objective Loss 0.041545 LR 0.000100 Time 0.558364
2022-11-29 00:26:44,810 - Epoch: [194][ 150/ 211] Overall Loss 0.041820 Objective Loss 0.041820 LR 0.000100 Time 0.555634
2022-11-29 00:26:49,948 - Epoch: [194][ 160/ 211] Overall Loss 0.041720 Objective Loss 0.041720 LR 0.000100 Time 0.553015
2022-11-29 00:26:55,135 - Epoch: [194][ 170/ 211] Overall Loss 0.041144 Objective Loss 0.041144 LR 0.000100 Time 0.550997
2022-11-29 00:27:00,327 - Epoch: [194][ 180/ 211] Overall Loss 0.041093 Objective Loss 0.041093 LR 0.000100 Time 0.549231
2022-11-29 00:27:05,496 - Epoch: [194][ 190/ 211] Overall Loss 0.040998 Objective Loss 0.040998 LR 0.000100 Time 0.547525
2022-11-29 00:27:10,677 - Epoch: [194][ 200/ 211] Overall Loss 0.040770 Objective Loss 0.040770 LR 0.000100 Time 0.546055
2022-11-29 00:27:15,845 - Epoch: [194][ 210/ 211] Overall Loss 0.040823 Objective Loss 0.040823 Top1 98.828125 Top5 100.000000 LR 0.000100 Time 0.544653
2022-11-29 00:27:16,335 - Epoch: [194][ 211/ 211] Overall Loss 0.040913 Objective Loss 0.040913 Top1 98.588710 Top5 100.000000 LR 0.000100 Time 0.544397
2022-11-29 00:27:16,871 - --- validate (epoch=194)-----------
2022-11-29 00:27:16,871 - 6000 samples (256 per mini-batch)
2022-11-29 00:27:25,000 - Epoch: [194][ 10/ 24] Loss 0.046200 Top1 98.789062 Top5 100.000000
2022-11-29 00:27:27,603 - Epoch: [194][ 20/ 24] Loss 0.041708 Top1 98.886719 Top5 100.000000
2022-11-29 00:27:28,506 - Epoch: [194][ 24/ 24] Loss 0.042684 Top1 98.833333 Top5 100.000000
2022-11-29 00:27:29,043 - ==> Top1: 98.833 Top5: 100.000 Loss: 0.043
2022-11-29 00:27:29,044 - ==> Confusion:
[[601 0 2 0 0 0 1 0 1 0]
[ 0 685 1 0 1 0 0 1 0 0]
[ 0 0 575 0 1 1 1 3 4 1]
[ 0 0 0 578 0 2 0 2 1 0]
[ 0 1 0 0 553 0 2 2 0 7]
[ 0 0 0 2 0 513 1 0 2 0]
[ 0 1 0 0 1 1 626 0 2 0]
[ 0 1 1 2 1 1 0 618 0 1]
[ 2 0 0 2 2 1 1 0 576 0]
[ 1 1 0 0 4 1 0 1 2 605]]
2022-11-29 00:27:29,046 - ==> Best [Top1: 99.067 Top5: 100.000 Sparsity:0.00 Params: 71148 on epoch: 173]
2022-11-29 00:27:29,046 - Saving checkpoint to: logs\2022.11.28-173243\qat_checkpoint.pth.tar
2022-11-29 00:27:29,050 -
2022-11-29 00:27:29,050 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-29 00:27:39,683 - Epoch: [195][ 10/ 211] Overall Loss 0.041090 Objective Loss 0.041090 LR 0.000100 Time 1.063157
2022-11-29 00:27:44,893 - Epoch: [195][ 20/ 211] Overall Loss 0.040325 Objective Loss 0.040325 LR 0.000100 Time 0.792032
2022-11-29 00:27:50,085 - Epoch: [195][ 30/ 211] Overall Loss 0.043324 Objective Loss 0.043324 LR 0.000100 Time 0.701092
2022-11-29 00:27:55,267 - Epoch: [195][ 40/ 211] Overall Loss 0.042096 Objective Loss 0.042096 LR 0.000100 Time 0.655373
2022-11-29 00:28:00,466 - Epoch: [195][ 50/ 211] Overall Loss 0.041351 Objective Loss 0.041351 LR 0.000100 Time 0.628260
2022-11-29 00:28:05,651 - Epoch: [195][ 60/ 211] Overall Loss 0.041290 Objective Loss 0.041290 LR 0.000100 Time 0.609952
2022-11-29 00:28:10,815 - Epoch: [195][ 70/ 211] Overall Loss 0.039647 Objective Loss 0.039647 LR 0.000100 Time 0.596576
2022-11-29 00:28:15,997 - Epoch: [195][ 80/ 211] Overall Loss 0.039195 Objective Loss 0.039195 LR 0.000100 Time 0.586781
2022-11-29 00:28:21,222 - Epoch: [195][ 90/ 211] Overall Loss 0.039482 Objective Loss 0.039482 LR 0.000100 Time 0.579628
2022-11-29 00:28:26,407 - Epoch: [195][ 100/ 211] Overall Loss 0.040421 Objective Loss 0.040421 LR 0.000100 Time 0.573506
2022-11-29 00:28:31,543 - Epoch: [195][ 110/ 211] Overall Loss 0.040668 Objective Loss 0.040668 LR 0.000100 Time 0.568045
2022-11-29 00:28:36,736 - Epoch: [195][ 120/ 211] Overall Loss 0.040550 Objective Loss 0.040550 LR 0.000100 Time 0.563984
2022-11-29 00:28:41,937 - Epoch: [195][ 130/ 211] Overall Loss 0.040587 Objective Loss 0.040587 LR 0.000100 Time 0.560609
2022-11-29 00:28:47,102 - Epoch: [195][ 140/ 211] Overall Loss 0.040201 Objective Loss 0.040201 LR 0.000100 Time 0.557445
2022-11-29 00:28:52,242 - Epoch: [195][ 150/ 211] Overall Loss 0.039808 Objective Loss 0.039808 LR 0.000100 Time 0.554537
2022-11-29 00:28:57,389 - Epoch: [195][ 160/ 211] Overall Loss 0.040046 Objective Loss 0.040046 LR 0.000100 Time 0.552036
2022-11-29 00:29:02,558 - Epoch: [195][ 170/ 211] Overall Loss 0.039700 Objective Loss 0.039700 LR 0.000100 Time 0.549958
2022-11-29 00:29:07,677 - Epoch: [195][ 180/ 211] Overall Loss 0.039700 Objective Loss 0.039700 LR 0.000100 Time 0.547835
2022-11-29 00:29:12,864 - Epoch: [195][ 190/ 211] Overall Loss 0.039723 Objective Loss 0.039723 LR 0.000100 Time 0.546297
2022-11-29 00:29:18,063 - Epoch: [195][ 200/ 211] Overall Loss 0.040026 Objective Loss 0.040026 LR 0.000100 Time 0.544972
2022-11-29 00:29:23,257 - Epoch: [195][ 210/ 211] Overall Loss 0.039709 Objective Loss 0.039709 Top1 99.218750 Top5 100.000000 LR 0.000100 Time 0.543755
2022-11-29 00:29:23,743 - Epoch: [195][ 211/ 211] Overall Loss 0.039742 Objective Loss 0.039742 Top1 99.193548 Top5 100.000000 LR 0.000100 Time 0.543475
2022-11-29 00:29:24,273 - --- validate (epoch=195)-----------
2022-11-29 00:29:24,273 - 6000 samples (256 per mini-batch)
2022-11-29 00:29:32,263 - Epoch: [195][ 10/ 24] Loss 0.032679 Top1 99.335938 Top5 100.000000
2022-11-29 00:29:34,916 - Epoch: [195][ 20/ 24] Loss 0.035417 Top1 99.199219 Top5 100.000000
2022-11-29 00:29:35,809 - Epoch: [195][ 24/ 24] Loss 0.038478 Top1 99.066667 Top5 100.000000
2022-11-29 00:29:36,337 - ==> Top1: 99.067 Top5: 100.000 Loss: 0.038
2022-11-29 00:29:36,337 - ==> Confusion:
[[600 0 1 1 0 0 1 0 0 2]
[ 0 684 2 0 0 0 0 2 0 0]
[ 0 0 582 1 0 0 0 1 1 1]
[ 0 0 1 578 0 1 0 2 0 1]
[ 0 2 0 0 557 0 0 1 0 5]
[ 1 1 0 2 1 511 2 0 0 0]
[ 0 0 0 0 1 0 630 0 0 0]
[ 0 2 1 0 0 0 0 622 0 0]
[ 0 0 1 3 0 2 0 1 577 0]
[ 0 2 1 0 3 1 0 2 3 603]]
2022-11-29 00:29:36,339 - ==> Best [Top1: 99.067 Top5: 100.000 Sparsity:0.00 Params: 71148 on epoch: 195]
2022-11-29 00:29:36,340 - Saving checkpoint to: logs\2022.11.28-173243\qat_checkpoint.pth.tar
2022-11-29 00:29:36,343 -
2022-11-29 00:29:36,343 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-29 00:29:46,980 - Epoch: [196][ 10/ 211] Overall Loss 0.043070 Objective Loss 0.043070 LR 0.000100 Time 1.063556
2022-11-29 00:29:52,160 - Epoch: [196][ 20/ 211] Overall Loss 0.040096 Objective Loss 0.040096 LR 0.000100 Time 0.790786
2022-11-29 00:29:57,371 - Epoch: [196][ 30/ 211] Overall Loss 0.040544 Objective Loss 0.040544 LR 0.000100 Time 0.700893
2022-11-29 00:30:02,562 - Epoch: [196][ 40/ 211] Overall Loss 0.042399 Objective Loss 0.042399 LR 0.000100 Time 0.655447
2022-11-29 00:30:07,759 - Epoch: [196][ 50/ 211] Overall Loss 0.042348 Objective Loss 0.042348 LR 0.000100 Time 0.628280
2022-11-29 00:30:12,935 - Epoch: [196][ 60/ 211] Overall Loss 0.041876 Objective Loss 0.041876 LR 0.000100 Time 0.609836
2022-11-29 00:30:18,117 - Epoch: [196][ 70/ 211] Overall Loss 0.041448 Objective Loss 0.041448 LR 0.000100 Time 0.596733
2022-11-29 00:30:23,360 - Epoch: [196][ 80/ 211] Overall Loss 0.041418 Objective Loss 0.041418 LR 0.000100 Time 0.587679
2022-11-29 00:30:28,616 - Epoch: [196][ 90/ 211] Overall Loss 0.041257 Objective Loss 0.041257 LR 0.000100 Time 0.580769
2022-11-29 00:30:33,810 - Epoch: [196][ 100/ 211] Overall Loss 0.040838 Objective Loss 0.040838 LR 0.000100 Time 0.574634
2022-11-29 00:30:39,097 - Epoch: [196][ 110/ 211] Overall Loss 0.040892 Objective Loss 0.040892 LR 0.000100 Time 0.570456
2022-11-29 00:30:44,332 - Epoch: [196][ 120/ 211] Overall Loss 0.041372 Objective Loss 0.041372 LR 0.000100 Time 0.566535
2022-11-29 00:30:49,530 - Epoch: [196][ 130/ 211] Overall Loss 0.041073 Objective Loss 0.041073 LR 0.000100 Time 0.562941
2022-11-29 00:30:54,714 - Epoch: [196][ 140/ 211] Overall Loss 0.040589 Objective Loss 0.040589 LR 0.000100 Time 0.559753
2022-11-29 00:30:59,888 - Epoch: [196][ 150/ 211] Overall Loss 0.040800 Objective Loss 0.040800 LR 0.000100 Time 0.556924
2022-11-29 00:31:05,017 - Epoch: [196][ 160/ 211] Overall Loss 0.040848 Objective Loss 0.040848 LR 0.000100 Time 0.554174
2022-11-29 00:31:10,209 - Epoch: [196][ 170/ 211] Overall Loss 0.040589 Objective Loss 0.040589 LR 0.000100 Time 0.552112
2022-11-29 00:31:15,420 - Epoch: [196][ 180/ 211] Overall Loss 0.040322 Objective Loss 0.040322 LR 0.000100 Time 0.550378
2022-11-29 00:31:20,649 - Epoch: [196][ 190/ 211] Overall Loss 0.040044 Objective Loss 0.040044 LR 0.000100 Time 0.548932
2022-11-29 00:31:25,829 - Epoch: [196][ 200/ 211] Overall Loss 0.040068 Objective Loss 0.040068 LR 0.000100 Time 0.547381
2022-11-29 00:31:31,004 - Epoch: [196][ 210/ 211] Overall Loss 0.039860 Objective Loss 0.039860 Top1 100.000000 Top5 100.000000 LR 0.000100 Time 0.545959
2022-11-29 00:31:31,495 - Epoch: [196][ 211/ 211] Overall Loss 0.039944 Objective Loss 0.039944 Top1 98.991935 Top5 100.000000 LR 0.000100 Time 0.545693
2022-11-29 00:31:32,036 - --- validate (epoch=196)-----------
2022-11-29 00:31:32,036 - 6000 samples (256 per mini-batch)
2022-11-29 00:31:40,151 - Epoch: [196][ 10/ 24] Loss 0.046594 Top1 98.632812 Top5 100.000000
2022-11-29 00:31:42,771 - Epoch: [196][ 20/ 24] Loss 0.044691 Top1 98.808594 Top5 100.000000
2022-11-29 00:31:43,685 - Epoch: [196][ 24/ 24] Loss 0.043413 Top1 98.816667 Top5 99.983333
2022-11-29 00:31:44,215 - ==> Top1: 98.817 Top5: 99.983 Loss: 0.043
2022-11-29 00:31:44,215 - ==> Confusion:
[[602 1 0 0 0 0 1 0 0 1]
[ 0 684 2 0 1 1 0 0 0 0]
[ 0 0 586 0 0 0 0 0 0 0]
[ 0 0 1 576 0 2 0 1 3 0]
[ 1 0 0 0 556 0 0 0 0 8]
[ 1 0 0 3 0 507 4 0 3 0]
[ 0 1 0 0 2 1 625 0 2 0]
[ 0 2 4 2 1 0 0 616 0 0]
[ 1 0 1 1 1 4 1 0 573 2]
[ 1 2 0 0 2 2 0 1 3 604]]
2022-11-29 00:31:44,217 - ==> Best [Top1: 99.067 Top5: 100.000 Sparsity:0.00 Params: 71148 on epoch: 195]
2022-11-29 00:31:44,218 - Saving checkpoint to: logs\2022.11.28-173243\qat_checkpoint.pth.tar
2022-11-29 00:31:44,220 -
2022-11-29 00:31:44,220 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-29 00:31:54,919 - Epoch: [197][ 10/ 211] Overall Loss 0.032844 Objective Loss 0.032844 LR 0.000100 Time 1.069839
2022-11-29 00:32:00,154 - Epoch: [197][ 20/ 211] Overall Loss 0.041889 Objective Loss 0.041889 LR 0.000100 Time 0.796520
2022-11-29 00:32:05,366 - Epoch: [197][ 30/ 211] Overall Loss 0.042166 Objective Loss 0.042166 LR 0.000100 Time 0.704649
2022-11-29 00:32:10,563 - Epoch: [197][ 40/ 211] Overall Loss 0.041843 Objective Loss 0.041843 LR 0.000100 Time 0.658390
2022-11-29 00:32:15,759 - Epoch: [197][ 50/ 211] Overall Loss 0.040065 Objective Loss 0.040065 LR 0.000100 Time 0.630634
2022-11-29 00:32:20,948 - Epoch: [197][ 60/ 211] Overall Loss 0.042284 Objective Loss 0.042284 LR 0.000100 Time 0.612014
2022-11-29 00:32:26,195 - Epoch: [197][ 70/ 211] Overall Loss 0.041101 Objective Loss 0.041101 LR 0.000100 Time 0.599525
2022-11-29 00:32:31,429 - Epoch: [197][ 80/ 211] Overall Loss 0.040270 Objective Loss 0.040270 LR 0.000100 Time 0.589997
2022-11-29 00:32:36,601 - Epoch: [197][ 90/ 211] Overall Loss 0.040462 Objective Loss 0.040462 LR 0.000100 Time 0.581911
2022-11-29 00:32:41,870 - Epoch: [197][ 100/ 211] Overall Loss 0.039729 Objective Loss 0.039729 LR 0.000100 Time 0.576399
2022-11-29 00:32:47,128 - Epoch: [197][ 110/ 211] Overall Loss 0.040009 Objective Loss 0.040009 LR 0.000100 Time 0.571798
2022-11-29 00:32:52,387 - Epoch: [197][ 120/ 211] Overall Loss 0.040253 Objective Loss 0.040253 LR 0.000100 Time 0.567965
2022-11-29 00:32:57,592 - Epoch: [197][ 130/ 211] Overall Loss 0.039340 Objective Loss 0.039340 LR 0.000100 Time 0.564314
2022-11-29 00:33:02,843 - Epoch: [197][ 140/ 211] Overall Loss 0.039311 Objective Loss 0.039311 LR 0.000100 Time 0.561499
2022-11-29 00:33:08,056 - Epoch: [197][ 150/ 211] Overall Loss 0.039410 Objective Loss 0.039410 LR 0.000100 Time 0.558812
2022-11-29 00:33:13,255 - Epoch: [197][ 160/ 211] Overall Loss 0.039413 Objective Loss 0.039413 LR 0.000100 Time 0.556381
2022-11-29 00:33:18,473 - Epoch: [197][ 170/ 211] Overall Loss 0.039128 Objective Loss 0.039128 LR 0.000100 Time 0.554335
2022-11-29 00:33:23,676 - Epoch: [197][ 180/ 211] Overall Loss 0.039173 Objective Loss 0.039173 LR 0.000100 Time 0.552445
2022-11-29 00:33:28,906 - Epoch: [197][ 190/ 211] Overall Loss 0.039230 Objective Loss 0.039230 LR 0.000100 Time 0.550895
2022-11-29 00:33:34,131 - Epoch: [197][ 200/ 211] Overall Loss 0.039786 Objective Loss 0.039786 LR 0.000100 Time 0.549471
2022-11-29 00:33:39,356 - Epoch: [197][ 210/ 211] Overall Loss 0.039715 Objective Loss 0.039715 Top1 98.046875 Top5 100.000000 LR 0.000100 Time 0.548187
2022-11-29 00:33:39,829 - Epoch: [197][ 211/ 211] Overall Loss 0.039615 Objective Loss 0.039615 Top1 98.991935 Top5 100.000000 LR 0.000100 Time 0.547829
2022-11-29 00:33:40,361 - --- validate (epoch=197)-----------
2022-11-29 00:33:40,361 - 6000 samples (256 per mini-batch)
2022-11-29 00:33:48,380 - Epoch: [197][ 10/ 24] Loss 0.034291 Top1 99.179688 Top5 100.000000
2022-11-29 00:33:51,013 - Epoch: [197][ 20/ 24] Loss 0.035368 Top1 99.179688 Top5 100.000000
2022-11-29 00:33:51,902 - Epoch: [197][ 24/ 24] Loss 0.036034 Top1 99.133333 Top5 100.000000
2022-11-29 00:33:52,430 - ==> Top1: 99.133 Top5: 100.000 Loss: 0.036
2022-11-29 00:33:52,431 - ==> Confusion:
[[603 0 1 0 0 0 1 0 0 0]
[ 0 687 0 0 1 0 0 0 0 0]
[ 0 1 575 2 1 0 0 5 1 1]
[ 0 0 2 577 0 2 0 1 1 0]
[ 0 0 0 0 560 0 0 0 0 5]
[ 1 0 0 1 0 513 2 0 1 0]
[ 1 1 0 0 2 0 625 0 2 0]
[ 0 0 1 1 0 0 0 623 0 0]
[ 0 0 0 1 0 0 4 0 577 2]
[ 1 1 0 1 2 1 0 0 1 608]]
2022-11-29 00:33:52,433 - ==> Best [Top1: 99.133 Top5: 100.000 Sparsity:0.00 Params: 71148 on epoch: 197]
2022-11-29 00:33:52,433 - Saving checkpoint to: logs\2022.11.28-173243\qat_checkpoint.pth.tar
2022-11-29 00:33:52,437 -
2022-11-29 00:33:52,437 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-29 00:34:03,037 - Epoch: [198][ 10/ 211] Overall Loss 0.040831 Objective Loss 0.040831 LR 0.000100 Time 1.059966
2022-11-29 00:34:08,190 - Epoch: [198][ 20/ 211] Overall Loss 0.039520 Objective Loss 0.039520 LR 0.000100 Time 0.787594
2022-11-29 00:34:13,378 - Epoch: [198][ 30/ 211] Overall Loss 0.038420 Objective Loss 0.038420 LR 0.000100 Time 0.698000
2022-11-29 00:34:18,509 - Epoch: [198][ 40/ 211] Overall Loss 0.040410 Objective Loss 0.040410 LR 0.000100 Time 0.651757
2022-11-29 00:34:23,625 - Epoch: [198][ 50/ 211] Overall Loss 0.041493 Objective Loss 0.041493 LR 0.000100 Time 0.623712
2022-11-29 00:34:28,804 - Epoch: [198][ 60/ 211] Overall Loss 0.040579 Objective Loss 0.040579 LR 0.000100 Time 0.606079
2022-11-29 00:34:33,965 - Epoch: [198][ 70/ 211] Overall Loss 0.040475 Objective Loss 0.040475 LR 0.000100 Time 0.593228
2022-11-29 00:34:39,143 - Epoch: [198][ 80/ 211] Overall Loss 0.040047 Objective Loss 0.040047 LR 0.000100 Time 0.583789
2022-11-29 00:34:44,312 - Epoch: [198][ 90/ 211] Overall Loss 0.039242 Objective Loss 0.039242 LR 0.000100 Time 0.576337
2022-11-29 00:34:49,498 - Epoch: [198][ 100/ 211] Overall Loss 0.039051 Objective Loss 0.039051 LR 0.000100 Time 0.570564
2022-11-29 00:34:54,671 - Epoch: [198][ 110/ 211] Overall Loss 0.038867 Objective Loss 0.038867 LR 0.000100 Time 0.565715
2022-11-29 00:34:59,814 - Epoch: [198][ 120/ 211] Overall Loss 0.038734 Objective Loss 0.038734 LR 0.000100 Time 0.561432
2022-11-29 00:35:04,991 - Epoch: [198][ 130/ 211] Overall Loss 0.038996 Objective Loss 0.038996 LR 0.000100 Time 0.558054
2022-11-29 00:35:10,177 - Epoch: [198][ 140/ 211] Overall Loss 0.039119 Objective Loss 0.039119 LR 0.000100 Time 0.555237
2022-11-29 00:35:15,315 - Epoch: [198][ 150/ 211] Overall Loss 0.038800 Objective Loss 0.038800 LR 0.000100 Time 0.552469
2022-11-29 00:35:20,470 - Epoch: [198][ 160/ 211] Overall Loss 0.038702 Objective Loss 0.038702 LR 0.000100 Time 0.550154
2022-11-29 00:35:25,639 - Epoch: [198][ 170/ 211] Overall Loss 0.038553 Objective Loss 0.038553 LR 0.000100 Time 0.548193
2022-11-29 00:35:30,769 - Epoch: [198][ 180/ 211] Overall Loss 0.038839 Objective Loss 0.038839 LR 0.000100 Time 0.546239
2022-11-29 00:35:35,934 - Epoch: [198][ 190/ 211] Overall Loss 0.039131 Objective Loss 0.039131 LR 0.000100 Time 0.544665
2022-11-29 00:35:41,090 - Epoch: [198][ 200/ 211] Overall Loss 0.039148 Objective Loss 0.039148 LR 0.000100 Time 0.543208
2022-11-29 00:35:46,284 - Epoch: [198][ 210/ 211] Overall Loss 0.038765 Objective Loss 0.038765 Top1 98.828125 Top5 100.000000 LR 0.000100 Time 0.542074
2022-11-29 00:35:46,767 - Epoch: [198][ 211/ 211] Overall Loss 0.038811 Objective Loss 0.038811 Top1 98.790323 Top5 100.000000 LR 0.000100 Time 0.541793
2022-11-29 00:35:47,300 - --- validate (epoch=198)-----------
2022-11-29 00:35:47,300 - 6000 samples (256 per mini-batch)
2022-11-29 00:35:55,284 - Epoch: [198][ 10/ 24] Loss 0.035620 Top1 99.101562 Top5 100.000000
2022-11-29 00:35:57,868 - Epoch: [198][ 20/ 24] Loss 0.042762 Top1 98.886719 Top5 100.000000
2022-11-29 00:35:58,761 - Epoch: [198][ 24/ 24] Loss 0.043785 Top1 98.850000 Top5 100.000000
2022-11-29 00:35:59,290 - ==> Top1: 98.850 Top5: 100.000 Loss: 0.044
2022-11-29 00:35:59,290 - ==> Confusion:
[[601 0 2 0 0 0 1 0 1 0]
[ 0 684 0 0 1 0 0 3 0 0]
[ 0 2 579 0 1 0 0 3 1 0]
[ 0 0 2 576 0 2 0 1 2 0]
[ 0 1 1 0 559 0 0 0 0 4]
[ 0 0 0 1 0 513 2 0 2 0]
[ 0 1 0 0 2 2 625 0 1 0]
[ 0 2 3 0 1 0 0 619 0 0]
[ 0 0 2 3 0 1 2 0 575 1]
[ 1 1 0 1 4 2 0 3 3 600]]
2022-11-29 00:35:59,292 - ==> Best [Top1: 99.133 Top5: 100.000 Sparsity:0.00 Params: 71148 on epoch: 197]
2022-11-29 00:35:59,293 - Saving checkpoint to: logs\2022.11.28-173243\qat_checkpoint.pth.tar
2022-11-29 00:35:59,295 -
2022-11-29 00:35:59,295 - Training epoch: 54000 samples (256 per mini-batch)
2022-11-29 00:36:09,907 - Epoch: [199][ 10/ 211] Overall Loss 0.049309 Objective Loss 0.049309 LR 0.000100 Time 1.061063
2022-11-29 00:36:15,067 - Epoch: [199][ 20/ 211] Overall Loss 0.045252 Objective Loss 0.045252 LR 0.000100 Time 0.788492
2022-11-29 00:36:20,240 - Epoch: [199][ 30/ 211] Overall Loss 0.043656 Objective Loss 0.043656 LR 0.000100 Time 0.698067
2022-11-29 00:36:25,401 - Epoch: [199][ 40/ 211] Overall Loss 0.044106 Objective Loss 0.044106 LR 0.000100 Time 0.652530
2022-11-29 00:36:30,563 - Epoch: [199][ 50/ 211] Overall Loss 0.045120 Objective Loss 0.045120 LR 0.000100 Time 0.625268
2022-11-29 00:36:35,687 - Epoch: [199][ 60/ 211] Overall Loss 0.043667 Objective Loss 0.043667 LR 0.000100 Time 0.606428
2022-11-29 00:36:40,813 - Epoch: [199][ 70/ 211] Overall Loss 0.043323 Objective Loss 0.043323 LR 0.000100 Time 0.593029
2022-11-29 00:36:45,963 - Epoch: [199][ 80/ 211] Overall Loss 0.044458 Objective Loss 0.044458 LR 0.000100 Time 0.583265
2022-11-29 00:36:51,112 - Epoch: [199][ 90/ 211] Overall Loss 0.043754 Objective Loss 0.043754 LR 0.000100 Time 0.575672
2022-11-29 00:36:56,255 - Epoch: [199][ 100/ 211] Overall Loss 0.043181 Objective Loss 0.043181 LR 0.000100 Time 0.569527
2022-11-29 00:37:01,418 - Epoch: [199][ 110/ 211] Overall Loss 0.043208 Objective Loss 0.043208 LR 0.000100 Time 0.564681
2022-11-29 00:37:06,581 - Epoch: [199][ 120/ 211] Overall Loss 0.042748 Objective Loss 0.042748 LR 0.000100 Time 0.560643
2022-11-29 00:37:11,725 - Epoch: [199][ 130/ 211] Overall Loss 0.042272 Objective Loss 0.042272 LR 0.000100 Time 0.557087
2022-11-29 00:37:16,852 - Epoch: [199][ 140/ 211] Overall Loss 0.042143 Objective Loss 0.042143 LR 0.000100 Time 0.553912
2022-11-29 00:37:22,042 - Epoch: [199][ 150/ 211] Overall Loss 0.042305 Objective Loss 0.042305 LR 0.000100 Time 0.551585
2022-11-29 00:37:27,176 - Epoch: [199][ 160/ 211] Overall Loss 0.041828 Objective Loss 0.041828 LR 0.000100 Time 0.549200
2022-11-29 00:37:32,331 - Epoch: [199][ 170/ 211] Overall Loss 0.041595 Objective Loss 0.041595 LR 0.000100 Time 0.547213
2022-11-29 00:37:37,568 - Epoch: [199][ 180/ 211] Overall Loss 0.041523 Objective Loss 0.041523 LR 0.000100 Time 0.545901
2022-11-29 00:37:42,684 - Epoch: [199][ 190/ 211] Overall Loss 0.041367 Objective Loss 0.041367 LR 0.000100 Time 0.544087
2022-11-29 00:37:47,835 - Epoch: [199][ 200/ 211] Overall Loss 0.041561 Objective Loss 0.041561 LR 0.000100 Time 0.542639
2022-11-29 00:37:52,985 - Epoch: [199][ 210/ 211] Overall Loss 0.041556 Objective Loss 0.041556 Top1 98.828125 Top5 100.000000 LR 0.000100 Time 0.541319
2022-11-29 00:37:53,473 - Epoch: [199][ 211/ 211] Overall Loss 0.041561 Objective Loss 0.041561 Top1 98.790323 Top5 100.000000 LR 0.000100 Time 0.541060
2022-11-29 00:37:54,003 - --- validate (epoch=199)-----------
2022-11-29 00:37:54,003 - 6000 samples (256 per mini-batch)
2022-11-29 00:38:01,956 - Epoch: [199][ 10/ 24] Loss 0.036670 Top1 98.906250 Top5 100.000000
2022-11-29 00:38:04,539 - Epoch: [199][ 20/ 24] Loss 0.039807 Top1 98.925781 Top5 100.000000
2022-11-29 00:38:05,419 - Epoch: [199][ 24/ 24] Loss 0.038310 Top1 98.950000 Top5 100.000000
2022-11-29 00:38:05,946 - ==> Top1: 98.950 Top5: 100.000 Loss: 0.038
2022-11-29 00:38:05,947 - ==> Confusion:
[[603 0 1 0 0 0 1 0 0 0]
[ 1 686 0 0 0 0 0 1 0 0]
[ 0 0 581 1 0 0 0 2 1 1]
[ 0 0 0 576 0 4 0 1 2 0]
[ 0 1 1 0 554 1 1 1 0 6]
[ 1 0 0 1 1 513 2 0 0 0]
[ 1 1 0 0 2 1 626 0 0 0]
[ 0 1 2 2 0 0 0 620 0 0]
[ 2 0 0 1 1 3 2 0 574 1]
[ 0 1 0 0 3 1 0 3 3 604]]
2022-11-29 00:38:05,949 - ==> Best [Top1: 99.133 Top5: 100.000 Sparsity:0.00 Params: 71148 on epoch: 197]
2022-11-29 00:38:05,949 - Saving checkpoint to: logs\2022.11.28-173243\qat_checkpoint.pth.tar
2022-11-29 00:38:05,951 - --- test ---------------------
2022-11-29 00:38:05,952 - 10000 samples (256 per mini-batch)
2022-11-29 00:38:13,849 - Test: [ 10/ 40] Loss 0.023113 Top1 99.257812 Top5 100.000000
2022-11-29 00:38:16,480 - Test: [ 20/ 40] Loss 0.022107 Top1 99.335938 Top5 100.000000
2022-11-29 00:38:19,102 - Test: [ 30/ 40] Loss 0.020887 Top1 99.401042 Top5 99.986979
2022-11-29 00:38:21,451 - Test: [ 40/ 40] Loss 0.020720 Top1 99.370000 Top5 99.990000
2022-11-29 00:38:21,984 - ==> Top1: 99.370 Top5: 99.990 Loss: 0.021
2022-11-29 00:38:21,984 - ==> Confusion:
[[ 976 0 1 1 0 0 2 0 0 0]
[ 0 1130 2 0 0 0 0 3 0 0]
[ 0 0 1028 0 0 0 0 3 1 0]
[ 0 0 0 1009 0 0 0 1 0 0]
[ 0 0 1 0 974 0 0 0 1 6]
[ 1 0 0 5 0 884 2 0 0 0]
[ 2 2 1 0 1 1 948 0 3 0]
[ 0 2 2 1 0 0 0 1022 0 1]
[ 0 0 3 1 1 1 0 0 966 2]
[ 0 0 0 0 5 1 0 2 1 1000]]
2022-11-29 00:38:21,993 -
2022-11-29 00:38:21,993 - Log file for this run: E:\MAX78000\ai8x-training\logs\2022.11.28-173243\2022.11.28-173243.log
最后的选择。最终决定放弃机器学习的任务，转而实现任务4。使用板卡上的摄像头完成以下一种或多种图像处理，并显示出处理后的图像：彩色图像噪点滤波，频域图像滤波，基于亮度的形态学处理（膨胀/腐蚀），图像背景估计，基于颜色的稀疏光流法。以前对图像的处理都是调用成熟的库。确实是很方便，但是对底层实现了解的不够清晰。趁这次机会好好学习一下图片卷积的应用，使用不同的卷积核就能实现图片的模糊、锐化、边缘检测灯功能。
任务实现：选定了任务，那就动手开始做吧！
首先是自己对卷积的理解。卷积算是图像处理中的基础操作，就如同使用不同的滤镜去观察图片，将需要的信息强化出来，不需要的信息弱化或抹去。这样图片指定的特征信息就变得明晰了，再交给后续处理。机器学习中卷积神经网络中的卷积也就是这么操作的。这里我实现了图片的边缘提取，实现了使用了Sobel 算子、Roberts算子、Prewitt 算子对摄像头图片进行边缘提取的功能。
1、读取图像。MAX78000开发板上集成了一颗摄像头，使用I2C通讯。直接参考例程ImgCapture，将摄像头的数据读取到内存了。这里读取到的是一个彩色的位图，图像格式为RGB565。卷积操作都是针对矩阵进行操作的，所以首先要将采集到的图片映射到二维矩阵中去。这里有两个选择，一种是将RGB564映射到红、绿、蓝三个二维矩阵中去，然后对每个矩阵进行卷积操作，最后再将三个矩阵合并成一幅图片。第二个方法就是将RGB的的图片转换为灰度图，这样就获得一个灰度图的二维矩阵，然后对灰度图进行卷积操作。这里我选择了第二个方法，使用灰度图进行操作。彩色图转灰度图使用公式：Gray = 0.299*R+0.587*G+0.114*B
// 从DMA中读取灰度图，写入内存
void read_grayimg_fromdma(uint32_t w, uint32_t h, uint8_t *imgdata) {
uint8_t *data = NULL;
uint16_t rgb;
uint16_t r, g, b;
// Get image line by line
for (int i = 0; i < h; i++) {
// Wait until camera streaming buffer is full
while ((data = get_camera_stream_buffer()) == NULL) {
if (camera_is_image_rcv()) {
break;
}
}
for (int j = 0; j < w; j++) {
rgb = data[j * 2] * 256 + data[j * 2 + 1];
r = (rgb & 0Xf800) >> 8;
g = (rgb & 0X07e0) >> 3;
b = (rgb & 0X001f) << 3;
imgdata[i * w + j] = (uint8_t) ((r * 30 + g * 59 + b * 11 + 50)
/ 100); //使用著名的心理学公式，转换为灰度图
// imgdata[i*w+j]=(((rgb & 0Xf800 )>>8)*30+((rgb & 0X07e0 )>>3)*59+((rgb & 0X001f )<<3)*11+50)/100;
}
// Release stream buffer
release_camera_stream_buffer();
}
// utils_stream_image_row_to_pc(imgdata, IMAGE_XRES * IMAGE_YRES * 2);
}
读取摄像头图片的尺寸，参考例程可以多种选择。但是在实际操作过程中，发现超过240X240后，貌似动态申请内存就会出错，暂时未能解决，这里就先使用240X240的图片尺寸。
当开始手写卷积操作，才痛苦地感觉到知易行难。各种和预想不一致的情况，为了乐能更好地测试卷积的算法，手写了一个测试案例图片，先用测试图片进行各种卷积操作。
//生成测试用图片
void read_grayimg_test(uint32_t w, uint32_t h, uint8_t *imgdata) {
for (int i = 0; i < h; i++) {
for (int j = 0; j < w; j++) {
imgdata[i * w + j] = 255 - (i / (h / 4)) * 64;
if (j > (w / 2 - 10) && j < (w / 2 + 10))
imgdata[i * w + j] = 255;
}
}
}2、卷积操作。这里选择的都是3X3的卷积核。依次取原图像中的每一个数据，并获取该图像数据和其四周的3X3数据，依次与卷积核对应数据相乘，再累加，作为原图像改位置新的数据。对原图像边缘场景，给与补零处理。
//卷积运算 kernel 是3*3的卷积核，边缘使用添0处理
void juanji(uint32_t w, uint32_t h, uint8_t *imgdata, uint8_t *dealdata) {
int sour[9]; //源数据
int sob_dx[9] = { 1, 0, -1, 2, 0, -2, 1, 0, -1 }; //Sobel 算子
int sob_dy[9] = { -1, -2, -1, 0, 0, 0, 1, 2, 1 };
int pre_dx[9] = { 1, 0, -1, 1, 0, -1, 1, 0, -1 }; //Prewitt 算子
int pre_dy[9] = { -1, -1, -1, 0, 0, 0, 1, 1, 1 };
int sch_dx[9] = { -3, 0, 3, -10, 0, 10, -3, 0, 3 }; //Scharr 算子
int sch_dy[9] = { -3, -10, -3, 0, 0, 0, 3, 10, 3 };
int newx, newy;
uint16_t x, y;
for (y = 0; y < h; y++) {
for (x = 0; x < w; x++) {
sour[4] = imgdata[y * w + x];
if (x == 0) {
sour[0] = 0;
sour[3] = 0;
sour[6] = 0;
sour[4] = imgdata[y * w + x];
sour[5] = imgdata[y * w + x + 1];
if (y == 0) {
sour[1] = 0;
sour[2] = 0;
sour[7] = imgdata[(y + 1) * w + x];
sour[8] = imgdata[(y + 1) * w + x + 1];
} else if (y == (h - 1)) {
sour[1] = imgdata[(y - 1) * w + x];
sour[2] = imgdata[(y - 1) * w + x + 1];
sour[7] = 0;
sour[8] = 0;
} else {
sour[1] = imgdata[(y - 1) * w + x];
sour[2] = imgdata[(y - 1) * w + x + 1];
sour[7] = imgdata[(y + 1) * w + x];
sour[8] = imgdata[(y + 1) * w + x + 1];
}
} else if (x == (w - 1)) {
sour[2] = 0;
sour[5] = 0;
sour[8] = 0;
sour[4] = imgdata[y * w + x];
sour[3] = imgdata[y * w + x - 1];
if (y == 0) {
sour[1] = 0;
sour[0] = 0;
sour[7] = imgdata[(y + 1) * w + x];
sour[6] = imgdata[(y + 1) * w + x - 1];
} else if (y == (h - 1)) {
sour[1] = imgdata[(y - 1) * w + x];
sour[0] = imgdata[(y - 1) * w + x - 1];
sour[7] = 0;
sour[6] = 0;
} else {
sour[1] = imgdata[(y - 1) * w + x];
sour[0] = imgdata[(y - 1) * w + x - 1];
sour[7] = imgdata[(y + 1) * w + x];
sour[6] = imgdata[(y + 1) * w + x - 1];
}
} else {
if (y == 0) {
sour[1] = 0;
sour[0] = 0;
sour[2] = 0;
sour[7] = imgdata[(y + 1) * w + x];
sour[6] = imgdata[(y + 1) * w + x - 1];
sour[8] = imgdata[(y + 1) * w + x + 1];
sour[3] = imgdata[y * w + x - 1];
sour[5] = imgdata[y * w + x + 1];
} else if (y == (h - 1)) {
sour[1] = imgdata[(y - 1) * w + x];
sour[0] = imgdata[(y - 1) * w + x - 1];
sour[3] = imgdata[y * w + x - 1];
sour[5] = imgdata[y * w + x + 1];
sour[7] = 0;
sour[6] = 0;
sour[8] = 0;
} else {
sour[1] = imgdata[(y - 1) * w + x];
sour[0] = imgdata[(y - 1) * w + x - 1];
sour[2] = imgdata[(y - 1) * w + x + 1];
sour[7] = imgdata[(y + 1) * w + x];
sour[6] = imgdata[(y + 1) * w + x - 1];
sour[8] = imgdata[(y + 1) * w + x + 1];
sour[3] = imgdata[y * w + x - 1];
sour[5] = imgdata[y * w + x + 1];
}
}
//矩阵乘
// dealdata[y * w + x] = matrixplus(sour, kernel);
if (button == 2) { //Prewitt 算子
newx = matrixplus(sour, pre_dx);
newy = matrixplus(sour, pre_dy);
}
if (button == 1) { //Scharr 算子
newx = matrixplus(sour, sch_dx);
newy = matrixplus(sour, sch_dy);
}
if (button == 0) { //Sobel 算子
newx = matrixplus(sour, sob_dx);
newy = matrixplus(sour, sob_dy);
}
newx = (int) sqrt(newx * newx + newy * newy);
// printf("\n%d %d %d\n",newx,newy,newxy);
if (newx > 255)
newx = 255;
if (newx < 0)
newx = 0;
dealdata[y * w + x] = newx;
}
}
}
对测试数据，可以看出Prewitt算子和Sobel算子结果差不多，Scharr算子有些区别。再看看真实环境的结果。
左侧均为实拍的灰度图。明显可以看出Prewitt算子在细微差别上边缘提取的更精致些。这里摘抄一下三个算子的特点：Sobel算子考虑了综合因素，对噪声较多的图像处理效果更好，Sobel 算子边缘定位效果不错，但检测出的边缘容易出现多像素宽度。Prewitt算子对灰度渐变的图像边缘提取效果较好，而没有考虑相邻点的距离远近对当前像素点的影响，与Sobel 算子类似，不同的是在平滑部分的权重大小有些差异。Scharr 算子:解决Sobel算子在梯度角度接近水平或垂直方向时的不精确性。Scharr通过将模版中的权重系数放大来增大像素值间的差异(计算 x 或 y 方向上的图像差分)。
后记：现在机器学习的功能，很多都在终端上实习开来。需要努力地学习这块的新知识！MAX78000板子还没能熟练掌握，获取图片速度慢，质量不高，例程中的CNN还不能为己所用，还有漫漫学习路要去走。感谢电子森林提供的这样的活动，及学习到了知识又开拓了眼界！
附件下载
CameraIF.zip
团队介绍
瞎捣鼓小能手
团队成员
aramy
单片机业余爱好者，瞎捣鼓小能手。
评论
0 / 100
查看更多
猜你喜欢
Max78000设计大赛中期报告--会伏丘的项目很有幸能够参加硬禾和ADI共同创办的本次智能边缘计算设计大赛，能够有机会在cortex内核中完成CNN模型的部署的尝试。截至目前已是中期，对前半段时间工作的总结整理如下。
会伏丘605
基于Max78000的人脸考勤机基于Max78000的人脸打卡机，训练使用阿里云抢占式服务器，底板提供了屏幕、WIFI、光源、电池电源等；设计了两个界面，分别是ESP8266配置界面和人脸识别界面。
退役熬夜选手1531
【MAX78000第二季】MAX78000FTHR-快速实现超低功耗、人工智能 (AI) 方案MAX78000FTHR 是一个快速开发平台，可帮助工程师使用 MAX78000 Arm®Cortex®快速实现超低功耗、人工智能 (AI) 解决方案，具有集成卷积神经网络加速器的 M4F 处理器。
Lucia5420
