FOTS.PyTorch
FOTS.PyTorch copied to clipboard
jiangxiluning
这个代码是完整的嘛？可以训练了嘛？Cannot compile lanms: /home/yana/OCR/FOTS.PyTorch/utils/lanms
那就是使用数据可以直接训练了，可是Cannot compile lanms: /home/yana/OCR/FOTS.PyTorch/utils/lanms这个是什么问题？GCC版本不够嘛，麻烦大牛了
感谢大神，已经在训练了，可是为什么验证的时候都为0 Train Epoch: 2 [0/804 (0%)] Loss: 4.113581 IOU Loss: 0.014848 CLS Loss: 0.009924 Recognition Loss: 4.088809 Train Epoch: 2 [36/804 (4%)] Loss: 3.984061 IOU Loss: 0.014813 CLS Loss: 0.009907 Recognition Loss: 3.959341 Train Epoch: 2 [72/804 (9%)] Loss: 4.084934 IOU Loss: 0.017774 CLS Loss: 0.009902 Recognition Loss: 4.057259 Train Epoch: 2 [108/804 (13%)] Loss: 3.899380 IOU Loss: 0.011619 CLS Loss: 0.009929 Recognition Loss: 3.877832 Train Epoch: 2 [144/804 (18%)] Loss: 4.338394 IOU Loss: 0.020665 CLS Loss: 0.009854 Recognition Loss: 4.307875 Train Epoch: 2 [180/804 (22%)] Loss: 3.913577 IOU Loss: 0.011609 CLS Loss: 0.009912 Recognition Loss: 3.892056 Train Epoch: 2 [216/804 (27%)] Loss: 3.734857 IOU Loss: 0.009591 CLS Loss: 0.009929 Recognition Loss: 3.715337 Train Epoch: 2 [252/804 (31%)] Loss: 4.268567 IOU Loss: 0.013152 CLS Loss: 0.009880 Recognition Loss: 4.245535 Train Epoch: 2 [288/804 (36%)] Loss: 4.136950 IOU Loss: 0.016697 CLS Loss: 0.009870 Recognition Loss: 4.110384 Train Epoch: 2 [324/804 (40%)] Loss: 4.046068 IOU Loss: 0.012804 CLS Loss: 0.009884 Recognition Loss: 4.023379 Train Epoch: 2 [360/804 (45%)] Loss: 4.105363 IOU Loss: 0.008889 CLS Loss: 0.009922 Recognition Loss: 4.086552 Train Epoch: 2 [396/804 (49%)] Loss: 4.104335 IOU Loss: 0.009707 CLS Loss: 0.009891 Recognition Loss: 4.084737 Train Epoch: 2 [432/804 (54%)] Loss: 3.972952 IOU Loss: 0.020410 CLS Loss: 0.009841 Recognition Loss: 3.942700 Train Epoch: 2 [468/804 (58%)] Loss: 4.076212 IOU Loss: 0.010874 CLS Loss: 0.009895 Recognition Loss: 4.055443 Train Epoch: 2 [504/804 (63%)] Loss: 4.043044 IOU Loss: 0.007563 CLS Loss: 0.009926 Recognition Loss: 4.025555 Train Epoch: 2 [540/804 (67%)] Loss: 4.177341 IOU Loss: 0.013657 CLS Loss: 0.009842 Recognition Loss: 4.153842 Train Epoch: 2 [576/804 (72%)] Loss: 4.024523 IOU Loss: 0.011835 CLS Loss: 0.009876 Recognition Loss: 4.002812 Train Epoch: 2 [612/804 (76%)] Loss: 3.715771 IOU Loss: 0.008423 CLS Loss: 0.009869 Recognition Loss: 3.697478 Train Epoch: 2 [648/804 (81%)] Loss: 3.883707 IOU Loss: 0.007087 CLS Loss: 0.009900 Recognition Loss: 3.866720 Train Epoch: 2 [684/804 (85%)] Loss: 3.857470 IOU Loss: 0.009716 CLS Loss: 0.009850 Recognition Loss: 3.837903 Train Epoch: 2 [720/804 (90%)] Loss: 3.985427 IOU Loss: 0.005914 CLS Loss: 0.009916 Recognition Loss: 3.969597 Train Epoch: 2 [756/804 (94%)] Loss: 3.975030 IOU Loss: 0.015484 CLS Loss: 0.009701 Recognition Loss: 3.949846 Train Epoch: 2 [792/804 (99%)] Loss: 4.167957 IOU Loss: 0.008214 CLS Loss: 0.009901 Recognition Loss: 4.149843 Cross point does not exist /home/yana/OCR/FOTS.PyTorch/FOTS/data_loader/datautils.py:230: RuntimeWarning: invalid value encountered in double_scalars return np.linalg.norm(np.cross(p2 - p1, p1 - p3)) / np.linalg.norm(p2 - p1) epoch : 2 loss : 4.008285451291212 precious : 0.0 recall : 0.0 hmean : 0.0 val_precious : 0.0 val_recall : 0.0 val_hmean : 0.0 Saving checkpoint: saved/FOT
@isyanan1024 i have the same problem with you , have you addressed it? @zhouyuangan could you please tell me where should i changed? you have completed the training?
@13438960761 It runs slowly but works actually. You can search how to change polygon coordinates formated with clockwise direction, choose PyimageSearch blog, which will help you find a good solution. The codes I won't share actually, I am sorry for that. Good luck.
@zhouyuangan Have you trained this model? I trained with two GPUs overnight and only trained 12% of one epoch.
Yes, I did, but it's very slow.
发自我的 iPad
------------------ 原始邮件 ------------------ 发件人: MiTeng0215 <[email protected]> 发送时间: 2020年2月4日 12:27 收件人: jiangxiluning/FOTS.PyTorch <[email protected]> 抄送: Yg Zhou <[email protected]>, Mention <[email protected]> 主题: 回复：[jiangxiluning/FOTS.PyTorch] 这个代码是完整的嘛？可以训练了嘛？Cannot compile lanms: /home/yana/OCR/FOTS.PyTorch/utils/lanms (#46)
@13438960761 It runs slowly but works actually. You can search how to change polygon coordinates formated with clockwise direction, choose PyimageSearch blog, which will help you find a good solution. The codes I won't share actually, I am sorry for that. Good luck.
Have you trained this model? I trained with two GPUs overnight and only trained 12% of one epoch.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.
How about the effect of detection and recognition？
发自我的 iPhone
在 2020年2月5日，17:08，Yg Zhou [email protected] 写道：
Yes, I did, but it's very slow.
发自我的 iPad
------------------ 原始邮件 ------------------ 发件人: MiTeng0215 <[email protected]> 发送时间: 2020年2月4日 12:27 收件人: jiangxiluning/FOTS.PyTorch <[email protected]> 抄送: Yg Zhou <[email protected]>, Mention <[email protected]> 主题: 回复：[jiangxiluning/FOTS.PyTorch] 这个代码是完整的嘛？可以训练了嘛？Cannot compile lanms: /home/yana/OCR/FOTS.PyTorch/utils/lanms (#46)
@13438960761 It runs slowly but works actually. You can search how to change polygon coordinates formated with clockwise direction, choose PyimageSearch blog, which will help you find a good solution. The codes I won't share actually, I am sorry for that. Good luck.
Have you trained this model? I trained with two GPUs overnight and only trained 12% of one epoch.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.
Traceback (most recent call last):
File "train.py", line 76, in
