zcswdt

Results 57 issues of zcswdt

![image](https://user-images.githubusercontent.com/43515926/108447654-359e5b80-729b-11eb-813c-00958cff358f.png)

模型在训练的时候前6个epoch都是耗时都是正常,4.5s左右。 ![image](https://user-images.githubusercontent.com/43515926/97249064-1425e980-183e-11eb-8bb1-7960335d44a9.png) 但是训练到第7个epoch的时候,10次迭代耗时450s。 ![image](https://user-images.githubusercontent.com/43515926/97248988-ec368600-183d-11eb-8975-4c959125b160.png)

大佬您好,我环境是双卡,15G显存,使用6万张图片,batch_size设置成8,numworks设置为0,能正常训练,设置其他的都要报错。请问这是什么原因导致的呢? ![image](https://user-images.githubusercontent.com/43515926/92883413-66a07780-f443-11ea-8653-17789c3f3282.png) ![image](https://user-images.githubusercontent.com/43515926/92883518-82a41900-f443-11ea-960b-91a9132b6eed.png)

感谢作者开源这个项目,我现在拿您的代码使用ic15,1c19,1c17和天猫数据训练了网络,100个epoch。准确召回如下: **recall: 0.586170, precision: 0.871295, hmean: 0.700843, train_loss: 0.759117。** 但是在检测的时候发现,文字框对于文字不能很好的包络, ![image](https://user-images.githubusercontent.com/43515926/89393155-febc8a00-d73c-11ea-9c9b-cda90a108ea9.png) 请问这种情况是是我的训练epoch不够,还是有可供调节的参数呢?期待您的回复。

https://github.com/backtime92/CRAFT-Reimplementation/blob/fbaa63aebd61c2b290752102cdef4758891b1fb7/gaussianMap/gaussian.py#L199 报错如下: cv2.error: OpenCV(4.2.0) C:\projects\opencv-python\opencv\modules\imgproc\src\colormap.cpp:714: error: (-5:Bad argument) cv::ColorMap only supports source images of type CV_8UC1 or CV_8UC3 in function 'cv::colormap::ColorMap::operator ()'

非常感谢你公开你的代码,看了您的实现,我有几个问题请教。 1.我看到您提供的预训练模型有用ic13+ic17训练的,但是我去看了这两个数据集的标注,ic13的gt是四个数字,ic17是8个数字,但是这两个数据集在数据加载的时候都是通过一个函数加载的。他们读gt坐标的时候不产生冲突吗? ![image](https://user-images.githubusercontent.com/43515926/86335506-2b363f80-bc81-11ea-9fee-19acc4e31300.png) 2.craft该方法是可以识别任意形状的文字框,我看您使用ic13,ic15,ic17,SynthText等数据集训练的,这些文字没有那种曲形(例如星巴克),但是训练出来模型也能识别这些形状的。能给个解释吗? 3.如果我想训练total_text数据集,这个代码能直接像ic15那样拿来训练吗?或者是我想加强模型对任意形状文字的检测能力,我该如何训练呢? 期待您的回复,谢谢。

I want to join the WeChat group, but the QR code has expired. Can I resend it or pull me into the group?

Excuse me, I saw that your code has been updated, and I re-use python main.py --full true now, but it produces thumbnails instead of full images. ![naver_0003](https://user-images.githubusercontent.com/43515926/137501615-16b354b1-a9f5-45a2-b95e-65f7a1bbe312.jpg) ![naver_0004](https://user-images.githubusercontent.com/43515926/137501688-7d2bbc73-80a2-4cba-a1ee-9d8fa95e6dd0.jpg) ![naver_0006](https://user-images.githubusercontent.com/43515926/137501708-1313b85f-6007-466f-bae9-271b48638563.jpg)

Hello, I can download thumbnails using `python main.py`, but can’t download with parameter `python main.py --full true`.Please help me, thanks! ![image](https://user-images.githubusercontent.com/43515926/104871970-8151a280-5987-11eb-9ca1-ff918f076478.png)

请问有没有公开的英文数据集?