Training with my own dataset

Bartzi / see

Code for the AAAI 2018 publication "SEE: Towards Semi-Supervised End-to-End Scene Text Recognition"

GNU General Public License v3.0

573 stars 147 forks source link

Training with my own dataset #87

Open muzaffersenkal opened 4 years ago

muzaffersenkal commented 4 years ago

Hi Bartzi ,

I am bothering you,sorry. I'm trying to recognize engraved text. So I created my own dataset.

I created ground truth file without box values. It looks like that ;

/see/new_dataset/1/1_ab12_12312.jpg    ab12
/see/new_dataset/1/2_cd52_12312.jpg    cd52

So is it enough for train? Can i train without box ? (I trained but accuracy looks 0 and loss value stucked about 0.48 )

Thank you

Bartzi commented 4 years ago

Hmm, looks good to me. You do not need any box annotation (that is the whole purpose of this research work).

I'm not sure why it does not work. Did you check everything that gets logged to the log_dir, especially the bboxes dir? This files there are intended to give you a feeling for the performance of the model over time.

muzaffersenkal commented 4 years ago

I understood my mistake. It's about gt file. I should be this format;

/see/new_dataset/1/1_ab12_12312.jpg 57 65 83 14

Now, it shows accuracy. but still accuracy too low. I guess I'm missing something. I'll examine deeper . Sorry for the unnecessary issue.