surfzjy / real-time-text-detection-dbnet Goto Github PK

I am trying to use pretrained model to train ResNet please give me some answer.
@SURFZJY Thank you.

File "/workspace/bin/base/base_trainer.py", line 104, in train

self._on_epoch_finish()
File "/workspace/bin/trainer/trainer.py", line 159, in _on_epoch_finish
recall, precision, hmean = self._eval()
File "/workspace/bin/trainer/trainer.py", line 144, in _eval
preds, boxes_list = decode(preds)
NameError: name 'decode' is not defined
main process return code: 0

在MacBook Pro，系统版本macOS 10.15.4，2.4GHz 四核 Intel Core i5，内存16G上，cpu推理，跑predict.py需要3.7s，这么慢的么，不能算real time了吧？

Thank you for your amazing code!!
I am a little confused about the input size of training data and predict data.
During training, the image size was set to (640, 640), while during predicting, the image size was set to (736, ) or (, 736), can you tell us why the input size are different? Thanks so much!!

It will be nice if you provide the link to weights not in baidu. I have struggled so much but finally got it .

Here is the link: https://mega.nz/#!WdhxXAxT!oGURvmbQFqTHu5hljUPdbDMzI75_UO2iWLaXX5dJrDw

Thanks for your work，your code is very clean. But the readme is too brief.
I wonder what dataset you used, and where should I put them.
Thank you again.

The trained model can be public? Is it consistent with the results in the paper?

Hello,
I was training and my training stop because of some reason but now I want to continue.

How can I download the datasets
ICDAR2013/15

Please explain for me about dilate text polygon at lines 78, 79. Why didn't you dilate it?
Thanks,
VietNV

@SURFZJY
How can I use your work if my dataset contains polygon instead of rectangles?

I am getting "load label failed" as of now or does it bound to have rectangle to use your work?

Hello
Why you use Dice_loss rather than BCE_Loss?
How to use BCE_loss?