search

argman / EAST

A tensorflow implementation of EAST text detector
GNU General Public License v3.0
3.02k stars 1.05k forks source link

Queries on training ? #72

Open anushaGundapaneni opened 6 years ago

anushaGundapaneni commented 6 years ago

I have following queries. Please do respond and provide some suggestions ::

1.should the image(jpg) and as well as (.txt) file should be there in the same folder or in different folders?

  1. how much time it takes to complete 1 step ?
  2. How many steps required to obtain better accuracy ?

Can you please help me with these questions !

argman commented 6 years ago

you can check the code in icdar.py this line see whether its reading annotation correctly.

anushaGundapaneni commented 6 years ago

Thanks for your response it is not reading annotation file. To get it read what i need to change in the code ?

can you please suggest ? screenshot from 2017-11-27 12-11-30

argman commented 6 years ago

check the function load_annotation to see whether your text format is correct, and you can write your own function to load that.

anushaGundapaneni commented 6 years ago

I'am unable to understand , Can you please eloberate . screenshot from 2017-11-28 09-30-01

In load_annotation it is csv.reader(f) , so, In order to load a text file do i need to change this line 55 ?

yingning commented 6 years ago

@anushaGundapaneni ,Have you solved this problem?

anushaGundapaneni commented 6 years ago

yes

The problem is Dataset is not loading !

In order to load that you need to change the file name of (gtimg.txt) to (img_.txt) and text files and images should be there in the same folder .

Anusha Gundapaneni Software Developer MobIcloud Infosystems Private Limited

On Thu, Jan 4, 2018 at 7:51 AM, yingning notifications@github.com wrote:

@anushaGundapaneni https://github.com/anushagundapaneni ,Have you solved this problem?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/argman/EAST/issues/72#issuecomment-355182504, or mute the thread https://github.com/notifications/unsubscribe-auth/Adn_oZpv55pwoMrpLgIyjcv53IP5tuA-ks5tHDWigaJpZM4QrHuF .

yingning commented 6 years ago

@anushaGundapaneni Thanks for your help .I have solved the loading problem .But After I trained my datasets ,I found my models weights performed poor in the test datasets.How does the "img_.txt" organized ?It is x1,y1,x2,y2,x3,y3,x4,y4, translated ?Does (x1,y1) means the left top corner of the rectangle

Vivianwxf commented 6 years ago

@yingning hello? I am confused that how can I make my own training dataset? Is there any tools or code preprocessing the samples? I am new to ICDAR， could you please help me? Also I have the same question with you, how do I understand the content of "img_*.txt"? Look forward to your response.