search

Bartzi / stn-ocr

Code for the paper STN-OCR: A single Neural Network for Text Detection and Text Recognition
https://arxiv.org/abs/1707.08831
GNU General Public License v3.0
499 stars 139 forks source link

Cannot train on fsns data set #5

Closed Jia-HongHenryLee closed 7 years ago

Jia-HongHenryLee commented 7 years ago

Hi Bartzi: i try to use train_fsns.py to train on fsns data set. i get the following error messages: if i use args, --eval-image
screenshot from 2017-09-29 16-23-47
i don't use args, --eval-image. i just load train_file and val_file screenshot from 2017-09-29 16-28-11

my development's environment: ubuntu 16.04 cuda 8.0 cudnn 5.0 mxnet 0.9.3

Thank you very much.

Bartzi commented 7 years ago

Hi,

I suspect there is something wrong with your annotation file. The shape of the softmax_labels should not be (128, 126) but rather something like (128, 21) or (128, 30)...

The error you get when not using an explicit eval image also contributes to this objection. Have a look at the annotation file you created, you might find the cause of the error there.

Jia-HongHenryLee commented 7 years ago

@Bartzi Thank you very much. I will try it.

Jia-HongHenryLee commented 7 years ago

@Bartzi Thank you very much. I execute the train_fsns.py successfully.

ThaiLe189 commented 4 years ago

Hi @Jia-HongHenryLee i get the same error how can you train_fnfs ? thank you !

Jia-HongHenryLee commented 4 years ago

Hi @fly809

You can follow the suggestion of @Bartzi to check annotation file and dimension of output layer.

ThaiLe189 commented 4 years ago

Hi @Jia-HongHenryLee Can you show me a example about a image and of label of it ? Thank you!