MichalBusta / DeepTextSpotter

285 stars 101 forks source link

failure in training #35

Open whulc opened 6 years ago

whulc commented 6 years ago
I try to reproduce your experiment on icdar dataset, but meet the following problem:  

qq 201805082215082 qq 201805082247052 qq 201805082247292 It seems that the data is not imported successfully, but I think the data is organized in the right way as required.

MichalBusta commented 6 years ago

maybe same issue as https://github.com/MichalBusta/DeepTextSpotter/issues/32 ?

whulc commented 6 years ago

Thank you, but I come up with another problem, I use the model you provide as pre-trained model, but in the training process, the predict result is right but it get a heavy loss, thus change the model vastly and the predict result get worse and worse. Here is a snapshot of the training process.

qq 20180514225528

phaertinger commented 6 years ago

Try index = 4 in train.py, line 46 (https://github.com/MichalBusta/DeepTextSpotter/blob/master/train.py#L46).

whulc commented 6 years ago

Thank you for your suggestion, but it doesn't work, the problem remains. tim 20180517171557

19931991 commented 6 years ago

hello , I research ocr recently , can we make a friend by qq:1323369151,thank you very much!@whulc

MichalBusta commented 6 years ago

Ref: #53 the problem:

mohammadSadeq commented 6 years ago

@MichalBusta @whulc I'm having the same issue(that the data is not imported successfully), how did you fix the problem after referencing issue #32?? Please do specify the things that you were doing wrong and what did you do to fix them Thank you

mohammadSadeq commented 6 years ago

@whulc Please help me with your first issue and how did you fix it, You will be a true life saver mate :D

MichalBusta commented 6 years ago

@MichalBusta @whulc I'm having the same issue(that the data is not imported successfully), how did you fix the problem after referencing issue #32??

the issue ref is #53

Please do specify the things that you were doing wrong and what did you do to fix them Thank you and again: solution is to pretrain stand-alone OCR on crops, and use model with optimizer parameters :)

mohammadSadeq commented 6 years ago

@MichalBusta I Believe you thought I was talking about the second problem, that is not my case, I'm getting this issue

I try to reproduce your experiment on icdar dataset, but meet the following problem:  

qq 201805082215082 qq 201805082247052 qq 201805082247292 It seems that the data is not imported successfully, but I think the data is organized in the right way as required.

lfdeep commented 6 years ago
I try to reproduce your experiment on icdar dataset, but meet the following problem:  

qq 201805082215082 qq 201805082247052 qq 201805082247292 It seems that the data is not imported successfully, but I think the data is organized in the right way as required.

Hello, i meet the same problem,How did you solve it?thanks