训练 xlvector的ocr和另一个类似的车牌识别始终精度为0，但作者训练是可以达到八九十精度的，新版mxnet有bug吗

[hopef]

一个是验证码ocr识别，作者文章介绍相当详细，表示可以达到98%：
https://github.com/xlvector/learning-dl/tree/master/mxnet/ocr

另一个是车牌识别，类似的，作者说达到了81%：
https://github.com/hopef/end-to-end-for-chinese-plate-recognition

可是我完整的跑下来，却两个案例始终是0，为什么啊，感觉是不是mxnet新版有bug啊

[winstywang]

You have to provide more details to diagnose the issues.

[winstywang]

@xlvector

[hopef]

非常感谢您的答复，我是在windows下win10 x64系统python跑的，测试了20160531_win2012_x64_gpu、20160419_win2012_x64_gpu、20160321_win2012_x64_gpu（该版本少了transpose symbol）共3个版本，下载自：https://github.com/dmlc/mxnet/releases 已经编译好的包包，使用GPU均完整跑完上面说的案例，以及使用CPU跑完xlvector的ocr案例，最终结果精度都是0（有时会是0.000110）

之所以怀疑是否windows版本的mxnet有bug，是因为我看博客上其他人也遇到类似的问题（连接找不到了），正准备去在ubuntu上测试看的，因为他们说成功的好像都是在linux下。

我还测试调大了数据量，比如从10万调整到了100万，学习率从0.001到0.01，以及batch_size从8调整到260等等，测试无数次没有一次成的，基本都是突然到0.003，然后慢慢掉回到0，我确认过期间产生的图片是正常的，一切看似都正常。同时我测试mxnet里面的案例比如nmist的结果也是正常，到底是什么缘故呢，谢谢！

以下是我刚刚跑的输出，代码一个没动，即使我无论怎么修改参数，结果也都类似：
Q:\Anaconda2\python.exe E:/dl/mxNet/workspace/my-ocr/cnn_ocr.py
2016-10-16 13:54:42,421 Start training with [gpu(0)]
2016-10-16 13:55:04,963 Epoch[0] Batch [50] Speed: 35.61 samples/sec Train-Accuracy=0.000000
2016-10-16 13:55:15,674 Epoch[0] Batch [100] Speed: 37.34 samples/sec Train-Accuracy=0.000000
2016-10-16 13:55:26,388 Epoch[0] Batch [150] Speed: 37.33 samples/sec Train-Accuracy=0.000000
2016-10-16 13:55:38,168 Epoch[0] Batch [200] Speed: 33.96 samples/sec Train-Accuracy=0.000000
2016-10-16 13:55:49,072 Epoch[0] Batch [250] Speed: 36.68 samples/sec Train-Accuracy=0.000000
2016-10-16 13:55:59,647 Epoch[0] Batch [300] Speed: 37.83 samples/sec Train-Accuracy=0.000000
2016-10-16 13:56:10,292 Epoch[0] Batch [350] Speed: 37.58 samples/sec Train-Accuracy=0.000000
2016-10-16 13:56:20,934 Epoch[0] Batch [400] Speed: 37.59 samples/sec Train-Accuracy=0.000000
2016-10-16 13:56:31,569 Epoch[0] Batch [450] Speed: 37.61 samples/sec Train-Accuracy=0.000000
2016-10-16 13:56:42,137 Epoch[0] Batch [500] Speed: 37.85 samples/sec Train-Accuracy=0.000000
2016-10-16 13:56:53,062 Epoch[0] Batch [550] Speed: 36.61 samples/sec Train-Accuracy=0.000000
2016-10-16 13:57:04,056 Epoch[0] Batch [600] Speed: 36.38 samples/sec Train-Accuracy=0.000000
2016-10-16 13:57:15,740 Epoch[0] Batch [650] Speed: 34.24 samples/sec Train-Accuracy=0.000000
2016-10-16 13:57:27,072 Epoch[0] Batch [700] Speed: 35.30 samples/sec Train-Accuracy=0.000000
2016-10-16 13:57:37,792 Epoch[0] Batch [750] Speed: 37.32 samples/sec Train-Accuracy=0.000000
2016-10-16 13:57:48,529 Epoch[0] Batch [800] Speed: 37.25 samples/sec Train-Accuracy=0.000000
2016-10-16 13:57:59,082 Epoch[0] Batch [850] Speed: 37.90 samples/sec Train-Accuracy=0.000000
2016-10-16 13:58:09,960 Epoch[0] Batch [900] Speed: 36.77 samples/sec Train-Accuracy=0.000000
2016-10-16 13:58:20,871 Epoch[0] Batch [950] Speed: 36.66 samples/sec Train-Accuracy=0.000000
2016-10-16 13:58:32,723 Epoch[0] Batch [1000] Speed: 33.75 samples/sec Train-Accuracy=0.000000
2016-10-16 13:58:44,013 Epoch[0] Batch [1050] Speed: 35.43 samples/sec Train-Accuracy=0.000000
2016-10-16 13:58:55,960 Epoch[0] Batch [1100] Speed: 33.48 samples/sec Train-Accuracy=0.000114
2016-10-16 13:59:07,135 Epoch[0] Batch [1150] Speed: 35.79 samples/sec Train-Accuracy=0.000109
2016-10-16 13:59:18,184 Epoch[0] Batch [1200] Speed: 36.22 samples/sec Train-Accuracy=0.000104
2016-10-16 13:59:29,164 Epoch[0] Batch [1250] Speed: 36.43 samples/sec Train-Accuracy=0.000100
2016-10-16 13:59:40,292 Epoch[0] Batch [1300] Speed: 35.96 samples/sec Train-Accuracy=0.000096
2016-10-16 13:59:51,341 Epoch[0] Batch [1350] Speed: 36.21 samples/sec Train-Accuracy=0.000093
2016-10-16 14:00:02,417 Epoch[0] Batch [1400] Speed: 36.11 samples/sec Train-Accuracy=0.000089
2016-10-16 14:00:13,599 Epoch[0] Batch [1450] Speed: 35.77 samples/sec Train-Accuracy=0.000086
2016-10-16 14:00:24,845 Epoch[0] Batch [1500] Speed: 35.57 samples/sec Train-Accuracy=0.000083
2016-10-16 14:00:35,529 Epoch[0] Batch [1550] Speed: 37.44 samples/sec Train-Accuracy=0.000081
2016-10-16 14:00:46,765 Epoch[0] Batch [1600] Speed: 35.60 samples/sec Train-Accuracy=0.000078
2016-10-16 14:00:58,673 Epoch[0] Batch [1650] Speed: 33.59 samples/sec Train-Accuracy=0.000076
2016-10-16 14:01:09,602 Epoch[0] Batch [1700] Speed: 36.60 samples/sec Train-Accuracy=0.000074
2016-10-16 14:01:20,693 Epoch[0] Batch [1750] Speed: 36.07 samples/sec Train-Accuracy=0.000071
2016-10-16 14:01:31,543 Epoch[0] Batch [1800] Speed: 36.86 samples/sec Train-Accuracy=0.000069
2016-10-16 14:01:42,374 Epoch[0] Batch [1850] Speed: 36.93 samples/sec Train-Accuracy=0.000068
2016-10-16 14:01:53,767 Epoch[0] Batch [1900] Speed: 35.11 samples/sec Train-Accuracy=0.000066
2016-10-16 14:02:04,687 Epoch[0] Batch [1950] Speed: 36.63 samples/sec Train-Accuracy=0.000064
2016-10-16 14:02:15,611 Epoch[0] Batch [2000] Speed: 36.62 samples/sec Train-Accuracy=0.000063
2016-10-16 14:02:26,907 Epoch[0] Batch [2050] Speed: 35.41 samples/sec Train-Accuracy=0.000061
2016-10-16 14:02:38,559 Epoch[0] Batch [2100] Speed: 34.34 samples/sec Train-Accuracy=0.000060
2016-10-16 14:02:50,102 Epoch[0] Batch [2150] Speed: 34.65 samples/sec Train-Accuracy=0.000058
2016-10-16 14:03:00,940 Epoch[0] Batch [2200] Speed: 36.91 samples/sec Train-Accuracy=0.000057
2016-10-16 14:03:12,061 Epoch[0] Batch [2250] Speed: 35.97 samples/sec Train-Accuracy=0.000056
2016-10-16 14:03:23,259 Epoch[0] Batch [2300] Speed: 35.72 samples/sec Train-Accuracy=0.000054
2016-10-16 14:03:34,193 Epoch[0] Batch [2350] Speed: 36.59 samples/sec Train-Accuracy=0.000053
2016-10-16 14:03:45,267 Epoch[0] Batch [2400] Speed: 36.12 samples/sec Train-Accuracy=0.000052
2016-10-16 14:03:56,384 Epoch[0] Batch [2450] Speed: 35.98 samples/sec Train-Accuracy=0.000051
2016-10-16 14:04:07,927 Epoch[0] Batch [2500] Speed: 34.66 samples/sec Train-Accuracy=0.000050
2016-10-16 14:04:18,779 Epoch[0] Batch [2550] Speed: 36.86 samples/sec Train-Accuracy=0.000049
2016-10-16 14:04:29,766 Epoch[0] Batch [2600] Speed: 36.41 samples/sec Train-Accuracy=0.000048
2016-10-16 14:04:40,654 Epoch[0] Batch [2650] Speed: 36.73 samples/sec Train-Accuracy=0.000094
2016-10-16 14:04:51,398 Epoch[0] Batch [2700] Speed: 37.23 samples/sec Train-Accuracy=0.000093
2016-10-16 14:05:03,102 Epoch[0] Batch [2750] Speed: 34.17 samples/sec Train-Accuracy=0.000091
2016-10-16 14:05:15,450 Epoch[0] Batch [2800] Speed: 32.39 samples/sec Train-Accuracy=0.000089
2016-10-16 14:05:26,835 Epoch[0] Batch [2850] Speed: 35.14 samples/sec Train-Accuracy=0.000088
2016-10-16 14:05:37,698 Epoch[0] Batch [2900] Speed: 36.82 samples/sec Train-Accuracy=0.000086
2016-10-16 14:05:48,759 Epoch[0] Batch [2950] Speed: 36.17 samples/sec Train-Accuracy=0.000085
2016-10-16 14:05:59,470 Epoch[0] Batch [3000] Speed: 37.34 samples/sec Train-Accuracy=0.000083
2016-10-16 14:06:10,187 Epoch[0] Batch [3050] Speed: 37.33 samples/sec Train-Accuracy=0.000082
2016-10-16 14:06:20,894 Epoch[0] Batch [3100] Speed: 37.36 samples/sec Train-Accuracy=0.000081
2016-10-16 14:06:31,592 Epoch[0] Batch [3150] Speed: 37.39 samples/sec Train-Accuracy=0.000079
2016-10-16 14:06:42,555 Epoch[0] Batch [3200] Speed: 36.49 samples/sec Train-Accuracy=0.000117
2016-10-16 14:06:53,131 Epoch[0] Batch [3250] Speed: 37.83 samples/sec Train-Accuracy=0.000115
2016-10-16 14:07:03,947 Epoch[0] Batch [3300] Speed: 36.98 samples/sec Train-Accuracy=0.000114
2016-10-16 14:07:15,009 Epoch[0] Batch [3350] Speed: 36.16 samples/sec Train-Accuracy=0.000112
2016-10-16 14:07:26,183 Epoch[0] Batch [3400] Speed: 35.80 samples/sec Train-Accuracy=0.000110
2016-10-16 14:07:36,937 Epoch[0] Batch [3450] Speed: 37.19 samples/sec Train-Accuracy=0.000109
2016-10-16 14:07:47,920 Epoch[0] Batch [3500] Speed: 36.42 samples/sec Train-Accuracy=0.000107
2016-10-16 14:07:58,563 Epoch[0] Batch [3550] Speed: 37.58 samples/sec Train-Accuracy=0.000106
2016-10-16 14:08:09,229 Epoch[0] Batch [3600] Speed: 37.51 samples/sec Train-Accuracy=0.000104
2016-10-16 14:08:19,759 Epoch[0] Batch [3650] Speed: 37.99 samples/sec Train-Accuracy=0.000103
2016-10-16 14:08:30,253 Epoch[0] Batch [3700] Speed: 38.11 samples/sec Train-Accuracy=0.000101
2016-10-16 14:08:41,559 Epoch[0] Batch [3750] Speed: 35.38 samples/sec Train-Accuracy=0.000100
2016-10-16 14:08:53,924 Epoch[0] Batch [3800] Speed: 32.35 samples/sec Train-Accuracy=0.000132
2016-10-16 14:09:05,388 Epoch[0] Batch [3850] Speed: 34.89 samples/sec Train-Accuracy=0.000130
2016-10-16 14:09:16,142 Epoch[0] Batch [3900] Speed: 37.19 samples/sec Train-Accuracy=0.000128
2016-10-16 14:09:27,046 Epoch[0] Batch [3950] Speed: 36.68 samples/sec Train-Accuracy=0.000158
2016-10-16 14:09:38,914 Epoch[0] Batch [4000] Speed: 33.70 samples/sec Train-Accuracy=0.000156
2016-10-16 14:09:50,390 Epoch[0] Batch [4050] Speed: 34.86 samples/sec Train-Accuracy=0.000154
2016-10-16 14:10:02,068 Epoch[0] Batch [4100] Speed: 34.25 samples/sec Train-Accuracy=0.000152
2016-10-16 14:10:13,315 Epoch[0] Batch [4150] Speed: 35.57 samples/sec Train-Accuracy=0.000151
2016-10-16 14:10:24,072 Epoch[0] Batch [4200] Speed: 37.19 samples/sec Train-Accuracy=0.000149
2016-10-16 14:10:35,302 Epoch[0] Batch [4250] Speed: 35.62 samples/sec Train-Accuracy=0.000147
2016-10-16 14:10:46,017 Epoch[0] Batch [4300] Speed: 37.33 samples/sec Train-Accuracy=0.000145
2016-10-16 14:10:58,246 Epoch[0] Batch [4350] Speed: 32.71 samples/sec Train-Accuracy=0.000144
2016-10-16 14:11:08,778 Epoch[0] Batch [4400] Speed: 37.98 samples/sec Train-Accuracy=0.000142
2016-10-16 14:11:19,240 Epoch[0] Batch [4450] Speed: 38.23 samples/sec Train-Accuracy=0.000169
2016-10-16 14:11:29,910 Epoch[0] Batch [4500] Speed: 37.49 samples/sec Train-Accuracy=0.000167
2016-10-16 14:11:40,594 Epoch[0] Batch [4550] Speed: 37.44 samples/sec Train-Accuracy=0.000165
2016-10-16 14:11:51,384 Epoch[0] Batch [4600] Speed: 37.07 samples/sec Train-Accuracy=0.000163
2016-10-16 14:12:02,084 Epoch[0] Batch [4650] Speed: 37.38 samples/sec Train-Accuracy=0.000161
2016-10-16 14:12:12,953 Epoch[0] Batch [4700] Speed: 36.80 samples/sec Train-Accuracy=0.000160
2016-10-16 14:12:23,698 Epoch[0] Batch [4750] Speed: 37.22 samples/sec Train-Accuracy=0.000184
2016-10-16 14:12:34,357 Epoch[0] Batch [4800] Speed: 37.58 samples/sec Train-Accuracy=0.000182
2016-10-16 14:12:45,108 Epoch[0] Batch [4850] Speed: 37.20 samples/sec Train-Accuracy=0.000180
2016-10-16 14:12:56,069 Epoch[0] Batch [4900] Speed: 36.50 samples/sec Train-Accuracy=0.000179
2016-10-16 14:13:07,680 Epoch[0] Batch [4950] Speed: 34.45 samples/sec Train-Accuracy=0.000177
2016-10-16 14:13:18,812 Epoch[0] Batch [5000] Speed: 35.94 samples/sec Train-Accuracy=0.000175
2016-10-16 14:13:29,709 Epoch[0] Batch [5050] Speed: 36.71 samples/sec Train-Accuracy=0.000173
2016-10-16 14:13:43,549 Epoch[0] Batch [5100] Speed: 28.90 samples/sec Train-Accuracy=0.000172
2016-10-16 14:14:01,207 Epoch[0] Batch [5150] Speed: 22.65 samples/sec Train-Accuracy=0.000170
2016-10-16 14:14:12,332 Epoch[0] Batch [5200] Speed: 35.95 samples/sec Train-Accuracy=0.000168
2016-10-16 14:14:23,078 Epoch[0] Batch [5250] Speed: 37.22 samples/sec Train-Accuracy=0.000167
2016-10-16 14:14:33,868 Epoch[0] Batch [5300] Speed: 37.07 samples/sec Train-Accuracy=0.000165
2016-10-16 14:14:45,506 Epoch[0] Batch [5350] Speed: 34.37 samples/sec Train-Accuracy=0.000164
2016-10-16 14:14:56,625 Epoch[0] Batch [5400] Speed: 35.97 samples/sec Train-Accuracy=0.000162
2016-10-16 14:15:07,729 Epoch[0] Batch [5450] Speed: 36.03 samples/sec Train-Accuracy=0.000161
2016-10-16 14:15:19,282 Epoch[0] Batch [5500] Speed: 34.62 samples/sec Train-Accuracy=0.000159
2016-10-16 14:15:30,171 Epoch[0] Batch [5550] Speed: 36.73 samples/sec Train-Accuracy=0.000158
2016-10-16 14:15:40,987 Epoch[0] Batch [5600] Speed: 36.98 samples/sec Train-Accuracy=0.000156
2016-10-16 14:15:52,367 Epoch[0] Batch [5650] Speed: 35.15 samples/sec Train-Accuracy=0.000155
2016-10-16 14:16:03,203 Epoch[0] Batch [5700] Speed: 36.91 samples/sec Train-Accuracy=0.000154
2016-10-16 14:16:14,299 Epoch[0] Batch [5750] Speed: 36.05 samples/sec Train-Accuracy=0.000152
2016-10-16 14:16:24,776 Epoch[0] Batch [5800] Speed: 38.18 samples/sec Train-Accuracy=0.000151
2016-10-16 14:16:35,551 Epoch[0] Batch [5850] Speed: 37.13 samples/sec Train-Accuracy=0.000150
2016-10-16 14:16:46,124 Epoch[0] Batch [5900] Speed: 37.83 samples/sec Train-Accuracy=0.000191
2016-10-16 14:16:56,790 Epoch[0] Batch [5950] Speed: 37.50 samples/sec Train-Accuracy=0.000189
2016-10-16 14:17:07,740 Epoch[0] Batch [6000] Speed: 36.53 samples/sec Train-Accuracy=0.000188
2016-10-16 14:17:18,552 Epoch[0] Batch [6050] Speed: 37.00 samples/sec Train-Accuracy=0.000186
2016-10-16 14:17:29,190 Epoch[0] Batch [6100] Speed: 37.60 samples/sec Train-Accuracy=0.000184
2016-10-16 14:17:40,305 Epoch[0] Batch [6150] Speed: 35.99 samples/sec Train-Accuracy=0.000183
2016-10-16 14:17:51,128 Epoch[0] Batch [6200] Speed: 36.95 samples/sec Train-Accuracy=0.000181
2016-10-16 14:18:02,338 Epoch[0] Batch [6250] Speed: 35.69 samples/sec Train-Accuracy=0.000180
2016-10-16 14:18:13,667 Epoch[0] Batch [6300] Speed: 35.30 samples/sec Train-Accuracy=0.000179
2016-10-16 14:18:25,042 Epoch[0] Batch [6350] Speed: 35.16 samples/sec Train-Accuracy=0.000177
2016-10-16 14:18:36,565 Epoch[0] Batch [6400] Speed: 34.71 samples/sec Train-Accuracy=0.000176
2016-10-16 14:18:47,602 Epoch[0] Batch [6450] Speed: 36.24 samples/sec Train-Accuracy=0.000174
2016-10-16 14:18:58,246 Epoch[0] Batch [6500] Speed: 37.58 samples/sec Train-Accuracy=0.000173
2016-10-16 14:19:08,930 Epoch[0] Batch [6550] Speed: 37.44 samples/sec Train-Accuracy=0.000172
2016-10-16 14:19:19,809 Epoch[0] Batch [6600] Speed: 36.77 samples/sec Train-Accuracy=0.000170
2016-10-16 14:19:30,454 Epoch[0] Batch [6650] Speed: 37.57 samples/sec Train-Accuracy=0.000169
2016-10-16 14:19:41,404 Epoch[0] Batch [6700] Speed: 36.53 samples/sec Train-Accuracy=0.000168
2016-10-16 14:19:52,322 Epoch[0] Batch [6750] Speed: 36.64 samples/sec Train-Accuracy=0.000167
2016-10-16 14:20:04,372 Epoch[0] Batch [6800] Speed: 33.20 samples/sec Train-Accuracy=0.000165
2016-10-16 14:20:15,121 Epoch[0] Batch [6850] Speed: 37.21 samples/sec Train-Accuracy=0.000164
2016-10-16 14:20:27,112 Epoch[0] Batch [6900] Speed: 33.36 samples/sec Train-Accuracy=0.000163
2016-10-16 14:20:38,790 Epoch[0] Batch [6950] Speed: 34.25 samples/sec Train-Accuracy=0.000162
2016-10-16 14:20:49,720 Epoch[0] Batch [7000] Speed: 36.60 samples/sec Train-Accuracy=0.000161
2016-10-16 14:21:01,187 Epoch[0] Batch [7050] Speed: 34.88 samples/sec Train-Accuracy=0.000160
2016-10-16 14:21:12,345 Epoch[0] Batch [7100] Speed: 35.85 samples/sec Train-Accuracy=0.000158
2016-10-16 14:21:23,109 Epoch[0] Batch [7150] Speed: 37.16 samples/sec Train-Accuracy=0.000157
2016-10-16 14:21:34,770 Epoch[0] Batch [7200] Speed: 34.30 samples/sec Train-Accuracy=0.000156
2016-10-16 14:21:46,250 Epoch[0] Batch [7250] Speed: 34.84 samples/sec Train-Accuracy=0.000155
2016-10-16 14:21:57,700 Epoch[0] Batch [7300] Speed: 34.93 samples/sec Train-Accuracy=0.000154
2016-10-16 14:22:08,848 Epoch[0] Batch [7350] Speed: 35.88 samples/sec Train-Accuracy=0.000153
2016-10-16 14:22:19,385 Epoch[0] Batch [7400] Speed: 37.96 samples/sec Train-Accuracy=0.000152
2016-10-16 14:22:30,157 Epoch[0] Batch [7450] Speed: 37.14 samples/sec Train-Accuracy=0.000151
2016-10-16 14:22:41,640 Epoch[0] Batch [7500] Speed: 34.83 samples/sec Train-Accuracy=0.000150
2016-10-16 14:22:53,025 Epoch[0] Batch [7550] Speed: 35.13 samples/sec Train-Accuracy=0.000149
2016-10-16 14:23:04,151 Epoch[0] Batch [7600] Speed: 35.95 samples/sec Train-Accuracy=0.000148
2016-10-16 14:23:15,223 Epoch[0] Batch [7650] Speed: 36.12 samples/sec Train-Accuracy=0.000147
2016-10-16 14:23:26,520 Epoch[0] Batch [7700] Speed: 35.41 samples/sec Train-Accuracy=0.000146
2016-10-16 14:23:37,346 Epoch[0] Batch [7750] Speed: 36.95 samples/sec Train-Accuracy=0.000145
2016-10-16 14:23:48,266 Epoch[0] Batch [7800] Speed: 36.63 samples/sec Train-Accuracy=0.000144
2016-10-16 14:23:59,401 Epoch[0] Batch [7850] Speed: 35.92 samples/sec Train-Accuracy=0.000143
2016-10-16 14:24:10,174 Epoch[0] Batch [7900] Speed: 37.13 samples/sec Train-Accuracy=0.000142
2016-10-16 14:24:21,039 Epoch[0] Batch [7950] Speed: 36.82 samples/sec Train-Accuracy=0.000142
2016-10-16 14:24:32,750 Epoch[0] Batch [8000] Speed: 34.16 samples/sec Train-Accuracy=0.000141
2016-10-16 14:24:43,957 Epoch[0] Batch [8050] Speed: 35.69 samples/sec Train-Accuracy=0.000140
2016-10-16 14:24:54,756 Epoch[0] Batch [8100] Speed: 37.04 samples/sec Train-Accuracy=0.000139
2016-10-16 14:25:05,727 Epoch[0] Batch [8150] Speed: 36.46 samples/sec Train-Accuracy=0.000138
2016-10-16 14:25:16,562 Epoch[0] Batch [8200] Speed: 36.92 samples/sec Train-Accuracy=0.000137
2016-10-16 14:25:27,746 Epoch[0] Batch [8250] Speed: 35.77 samples/sec Train-Accuracy=0.000136
2016-10-16 14:25:38,733 Epoch[0] Batch [8300] Speed: 36.41 samples/sec Train-Accuracy=0.000136
2016-10-16 14:25:49,651 Epoch[0] Batch [8350] Speed: 36.64 samples/sec Train-Accuracy=0.000135
2016-10-16 14:26:00,509 Epoch[0] Batch [8400] Speed: 36.84 samples/sec Train-Accuracy=0.000134
2016-10-16 14:26:11,196 Epoch[0] Batch [8450] Speed: 37.43 samples/sec Train-Accuracy=0.000133
2016-10-16 14:26:22,009 Epoch[0] Batch [8500] Speed: 36.99 samples/sec Train-Accuracy=0.000132
2016-10-16 14:26:32,766 Epoch[0] Batch [8550] Speed: 37.18 samples/sec Train-Accuracy=0.000132
2016-10-16 14:26:43,509 Epoch[0] Batch [8600] Speed: 37.24 samples/sec Train-Accuracy=0.000131
2016-10-16 14:26:54,230 Epoch[0] Batch [8650] Speed: 37.31 samples/sec Train-Accuracy=0.000130
2016-10-16 14:27:04,954 Epoch[0] Batch [8700] Speed: 37.30 samples/sec Train-Accuracy=0.000129
2016-10-16 14:27:16,040 Epoch[0] Batch [8750] Speed: 36.08 samples/sec Train-Accuracy=0.000129
2016-10-16 14:27:26,933 Epoch[0] Batch [8800] Speed: 36.72 samples/sec Train-Accuracy=0.000128
2016-10-16 14:27:38,073 Epoch[0] Batch [8850] Speed: 35.96 samples/sec Train-Accuracy=0.000127
2016-10-16 14:27:49,289 Epoch[0] Batch [8900] Speed: 35.67 samples/sec Train-Accuracy=0.000126
2016-10-16 14:28:00,066 Epoch[0] Batch [8950] Speed: 37.11 samples/sec Train-Accuracy=0.000126
2016-10-16 14:28:11,292 Epoch[0] Batch [9000] Speed: 35.63 samples/sec Train-Accuracy=0.000125
2016-10-16 14:28:22,788 Epoch[0] Batch [9050] Speed: 34.80 samples/sec Train-Accuracy=0.000124
2016-10-16 14:28:33,799 Epoch[0] Batch [9100] Speed: 36.33 samples/sec Train-Accuracy=0.000124
2016-10-16 14:28:44,494 Epoch[0] Batch [9150] Speed: 37.40 samples/sec Train-Accuracy=0.000123
2016-10-16 14:28:55,220 Epoch[0] Batch [9200] Speed: 37.29 samples/sec Train-Accuracy=0.000122
2016-10-16 14:29:05,836 Epoch[0] Batch [9250] Speed: 37.68 samples/sec Train-Accuracy=0.000122
2016-10-16 14:29:16,592 Epoch[0] Batch [9300] Speed: 37.19 samples/sec Train-Accuracy=0.000121
2016-10-16 14:29:27,272 Epoch[0] Batch [9350] Speed: 37.45 samples/sec Train-Accuracy=0.000120
2016-10-16 14:29:37,904 Epoch[0] Batch [9400] Speed: 37.62 samples/sec Train-Accuracy=0.000120
2016-10-16 14:29:48,713 Epoch[0] Batch [9450] Speed: 37.00 samples/sec Train-Accuracy=0.000119
2016-10-16 14:29:59,397 Epoch[0] Batch [9500] Speed: 37.44 samples/sec Train-Accuracy=0.000118
2016-10-16 14:30:10,220 Epoch[0] Batch [9550] Speed: 36.96 samples/sec Train-Accuracy=0.000118
2016-10-16 14:30:20,767 Epoch[0] Batch [9600] Speed: 37.92 samples/sec Train-Accuracy=0.000117
2016-10-16 14:30:31,509 Epoch[0] Batch [9650] Speed: 37.24 samples/sec Train-Accuracy=0.000117
2016-10-16 14:30:42,098 Epoch[0] Batch [9700] Speed: 37.77 samples/sec Train-Accuracy=0.000116
2016-10-16 14:30:53,390 Epoch[0] Batch [9750] Speed: 35.43 samples/sec Train-Accuracy=0.000115
2016-10-16 14:31:04,255 Epoch[0] Batch [9800] Speed: 36.82 samples/sec Train-Accuracy=0.000115
2016-10-16 14:31:14,983 Epoch[0] Batch [9850] Speed: 37.29 samples/sec Train-Accuracy=0.000114
2016-10-16 14:31:25,710 Epoch[0] Batch [9900] Speed: 37.29 samples/sec Train-Accuracy=0.000114
2016-10-16 14:31:36,428 Epoch[0] Batch [9950] Speed: 37.32 samples/sec Train-Accuracy=0.000113
2016-10-16 14:31:47,187 Epoch[0] Batch [10000] Speed: 37.18 samples/sec Train-Accuracy=0.000112
2016-10-16 14:31:57,947 Epoch[0] Batch [10050] Speed: 37.23 samples/sec Train-Accuracy=0.000112
2016-10-16 14:32:08,612 Epoch[0] Batch [10100] Speed: 37.51 samples/sec Train-Accuracy=0.000111
2016-10-16 14:32:19,479 Epoch[0] Batch [10150] Speed: 36.81 samples/sec Train-Accuracy=0.000111
2016-10-16 14:32:30,217 Epoch[0] Batch [10200] Speed: 37.25 samples/sec Train-Accuracy=0.000110
2016-10-16 14:32:41,089 Epoch[0] Batch [10250] Speed: 36.79 samples/sec Train-Accuracy=0.000110
2016-10-16 14:32:51,674 Epoch[0] Batch [10300] Speed: 37.79 samples/sec Train-Accuracy=0.000109
2016-10-16 14:33:02,372 Epoch[0] Batch [10350] Speed: 37.39 samples/sec Train-Accuracy=0.000109
2016-10-16 14:33:13,255 Epoch[0] Batch [10400] Speed: 36.76 samples/sec Train-Accuracy=0.000108
2016-10-16 14:33:24,250 Epoch[0] Batch [10450] Speed: 36.38 samples/sec Train-Accuracy=0.000108
2016-10-16 14:33:35,430 Epoch[0] Batch [10500] Speed: 35.78 samples/sec Train-Accuracy=0.000107
2016-10-16 14:33:46,749 Epoch[0] Batch [10550] Speed: 35.34 samples/sec Train-Accuracy=0.000107
2016-10-16 14:33:57,788 Epoch[0] Batch [10600] Speed: 36.24 samples/sec Train-Accuracy=0.000106
2016-10-16 14:34:09,398 Epoch[0] Batch [10650] Speed: 34.45 samples/sec Train-Accuracy=0.000106
2016-10-16 14:34:20,713 Epoch[0] Batch [10700] Speed: 35.35 samples/sec Train-Accuracy=0.000105
2016-10-16 14:34:31,743 Epoch[0] Batch [10750] Speed: 36.26 samples/sec Train-Accuracy=0.000105
2016-10-16 14:34:42,329 Epoch[0] Batch [10800] Speed: 37.79 samples/sec Train-Accuracy=0.000104
2016-10-16 14:34:53,048 Epoch[0] Batch [10850] Speed: 37.32 samples/sec Train-Accuracy=0.000104
2016-10-16 14:35:03,931 Epoch[0] Batch [10900] Speed: 36.75 samples/sec Train-Accuracy=0.000103
2016-10-16 14:35:14,671 Epoch[0] Batch [10950] Speed: 37.25 samples/sec Train-Accuracy=0.000114
2016-10-16 14:35:25,415 Epoch[0] Batch [11000] Speed: 37.23 samples/sec Train-Accuracy=0.000114
2016-10-16 14:35:36,220 Epoch[0] Batch [11050] Speed: 37.02 samples/sec Train-Accuracy=0.000113
2016-10-16 14:35:46,927 Epoch[0] Batch [11100] Speed: 37.36 samples/sec Train-Accuracy=0.000113
2016-10-16 14:35:58,072 Epoch[0] Batch [11150] Speed: 35.89 samples/sec Train-Accuracy=0.000112
2016-10-16 14:36:09,099 Epoch[0] Batch [11200] Speed: 36.27 samples/sec Train-Accuracy=0.000112
2016-10-16 14:36:20,707 Epoch[0] Batch [11250] Speed: 34.46 samples/sec Train-Accuracy=0.000111
2016-10-16 14:36:31,849 Epoch[0] Batch [11300] Speed: 35.90 samples/sec Train-Accuracy=0.000111
2016-10-16 14:36:42,770 Epoch[0] Batch [11350] Speed: 36.63 samples/sec Train-Accuracy=0.000110
2016-10-16 14:36:53,803 Epoch[0] Batch [11400] Speed: 36.25 samples/sec Train-Accuracy=0.000110
2016-10-16 14:37:04,858 Epoch[0] Batch [11450] Speed: 36.18 samples/sec Train-Accuracy=0.000109
2016-10-16 14:37:15,688 Epoch[0] Batch [11500] Speed: 36.93 samples/sec Train-Accuracy=0.000109
2016-10-16 14:37:26,460 Epoch[0] Batch [11550] Speed: 37.14 samples/sec Train-Accuracy=0.000108
2016-10-16 14:37:37,177 Epoch[0] Batch [11600] Speed: 37.32 samples/sec Train-Accuracy=0.000108
2016-10-16 14:37:47,782 Epoch[0] Batch [11650] Speed: 37.72 samples/sec Train-Accuracy=0.000107
2016-10-16 14:37:58,322 Epoch[0] Batch [11700] Speed: 37.95 samples/sec Train-Accuracy=0.000107
2016-10-16 14:38:09,203 Epoch[0] Batch [11750] Speed: 36.76 samples/sec Train-Accuracy=0.000106
2016-10-16 14:38:19,736 Epoch[0] Batch [11800] Speed: 37.98 samples/sec Train-Accuracy=0.000106
2016-10-16 14:38:30,424 Epoch[0] Batch [11850] Speed: 37.43 samples/sec Train-Accuracy=0.000116
2016-10-16 14:38:42,023 Epoch[0] Batch [11900] Speed: 34.49 samples/sec Train-Accuracy=0.000116
2016-10-16 14:38:53,012 Epoch[0] Batch [11950] Speed: 36.40 samples/sec Train-Accuracy=0.000115
2016-10-16 14:39:03,585 Epoch[0] Batch [12000] Speed: 37.83 samples/sec Train-Accuracy=0.000115
2016-10-16 14:39:14,313 Epoch[0] Batch [12050] Speed: 37.28 samples/sec Train-Accuracy=0.000114
2016-10-16 14:39:25,137 Epoch[0] Batch [12100] Speed: 36.96 samples/sec Train-Accuracy=0.000114
2016-10-16 14:39:35,864 Epoch[0] Batch [12150] Speed: 37.29 samples/sec Train-Accuracy=0.000113
2016-10-16 14:39:46,562 Epoch[0] Batch [12200] Speed: 37.39 samples/sec Train-Accuracy=0.000113
2016-10-16 14:39:57,369 Epoch[0] Batch [12250] Speed: 37.01 samples/sec Train-Accuracy=0.000112
2016-10-16 14:40:07,927 Epoch[0] Batch [12300] Speed: 37.89 samples/sec Train-Accuracy=0.000112
2016-10-16 14:40:18,559 Epoch[0] Batch [12350] Speed: 37.62 samples/sec Train-Accuracy=0.000111
2016-10-16 14:40:29,381 Epoch[0] Batch [12400] Speed: 36.96 samples/sec Train-Accuracy=0.000111
2016-10-16 14:40:40,127 Epoch[0] Batch [12450] Speed: 37.23 samples/sec Train-Accuracy=0.000110
2016-10-16 14:40:50,805 Epoch[0] Batch [12500] Speed: 37.46 samples/sec Train-Accuracy=0.000110
2016-10-16 14:40:50,805 Epoch[0] Resetting Data Iterator
2016-10-16 14:40:50,805 Epoch[0] Train-Accuracy=0.000110
2016-10-16 14:40:50,805 Epoch[0] Time cost=2764.878
2016-10-16 14:40:57,308 Epoch[0] Validation-Accuracy=0.000000
2016-10-16 14:40:57,496 Saved checkpoint to "cnn-ocr-0001.params"

Process finished with exit code 0

[xlvector]

cnn的不应该啊。图片一开始所有的像素除以255了吗？

[hopef]

@xlvector 除以了，我是用您的代码，一分一毫都没有修改跑的
https://github.com/xlvector/learning-dl/blob/master/mxnet/ocr/cnn_ocr.py
而且我加大了数据量测试依旧不行，win10、win7下都不行

[hopef]

我看到这里同样提到这种情况：
xlvector/learning-dl#6
难道真的是mxnet的windows版本有bug吗？

[xlvector]

看了最简单的方法是去aws申请一台linux的机器，同样代码跑一遍看看。。。

[hopef]

我已经在尝试装linux版本的mxnet测试，只是今天一天装系统的过程不顺畅，没有完成测试。一旦测试完成第一时间在这里答复。

[yajiedesign]

you version too old.
you can download new version in https://github.com/yajiedesign/mxnet/releases
my build new version every day.

[hopef]

@yajiedesign hahhaha，是我的version too old。尝试您编译的，结果正常的很，这里的https://github.com/dmlc/mxnet/releases 编译的windows版本就是坑爹的坑货，我真真是测试了无数遍呀。。。 灰常感谢您，谢谢~

[thirdwing]

@yajiedesign I will also use your daily release for the R package on Windows. Is it also possible to provide CPU-only version?

[yajiedesign]

@thirdwing I'll try to offer later.

[csJoax]

我也碰到过类似的情况，用的是ubuntu系统下kaixhin/mxnet的docker镜像。到了第20个epoch，测试精度很低且没有任何变化。
PS：使用CPU计算；用的是python接口；网络模型是在Alexnet的基础上做了很小的修改（卷积核11x11—>10x10）；数据集倒不是很权威，但也算可靠。

[yajiedesign]

@thirdwing sorry some late.cpu version has been uploaded

[thirdwing]

@yajiedesign Thanks!

BTW, I think we can close this issue for now.