search

huoyijie / AdvancedEAST

AdvancedEAST is an algorithm used for Scene image text detect, which is primarily based on EAST, and the significant improvement was also made, which make long text predictions more accurate.https://github.com/huoyijie/raspberrypi-car
https://huoyijie.cn/
MIT License
1.23k stars 380 forks source link

Accurate of AdvancedEAST in ICDAR2015 is not good #64

Open hcnhatnam opened 5 years ago

hcnhatnam commented 5 years ago

I get pretrained model in https://github.com/gfickel/advanced_east_weights(provided on issue #40) and resul is not good Screenshot from 2019-03-23 14-10-38. Can anyone explain this problem?

aiguoth commented 5 years ago

@hcnhatnam hi, can you provide the website for this test? I want to test my model ,thank you.

hcnhatnam commented 5 years ago

@Guoth Focused Scene Text 2013-2015(icdar2015) Screenshot from 2019-03-23 21-42-31 you can download Offline evaluation(this is a evaluation tool).

aiguoth commented 5 years ago

Thank you very much for your help. Do you have any test sets about ICDAR2013 and ICDAR2011. I try to register on the official website : http://www.cvc.uab.es/icdar2011competition/?com=contestant , but I didn't receive an email about the password, so I couldn't download the test sets. if you have these test sets about ICDAR2013 and ICDAR2011,could you sent it to me.I'll be very grateful for your help ------------------ Original ------------------ From: "hcnhatnam"notifications@github.com; Date: Sat, Mar 23, 2019 10:43 PM To: "huoyijie/AdvancedEAST"AdvancedEAST@noreply.github.com; Cc: "777777777"guoth@smail.nju.edu.cn; "Mention"mention@noreply.github.com; Subject: Re: [huoyijie/AdvancedEAST] Accurate of AdvancedEAST in ICDAR2015 is not good (#64)

@Guoth Focused Scene Text 2013-2015(icdar2015)

you canan dowload Offline evaluation(this is a evaluation tool).

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

hcnhatnam commented 5 years ago

you need to 1 account to access into dataset and submit. you can register in here.

aiguoth commented 5 years ago

@hcnhatnam thank you very much for your help.

hcnhatnam commented 5 years ago

@hcnhatnam thank you very much for your help.

@Guoth Did yout test your model? What is the result?

aiguoth commented 5 years ago

@hcnhatnam I used this repo to test the ICDAR2015 test set, but only 359 images output coordinate information, so I would like to ask how you tested the accuracy results. I don't get the result by myself.

hcnhatnam commented 5 years ago

@Guoth Can you share model for me,then i get the result for you?email:nhatnam10a3@gmail.com

THUHJ commented 5 years ago

@hcnhatnam hi, I also tried the pre-trained model in ( https://github.com/gfickel/advanced_east_weights) on icdar 2015. My Hmeans is only around 30%.

Did you change the threshold in the cfg.py?

hcnhatnam commented 5 years ago

30%.. Really, no, i don't change and i used to model in this model to test

THUHJ commented 5 years ago

....I use the same model.....;Sad And I also tried retrain the model based on the icdar 2015 data but the result is also around 30%~40%.... Maybe I should clean anything and try again....

THUHJ commented 5 years ago

@hcnhatnam Have you tried fine-tune the pre-trained model based on icdar 2015 data?

hcnhatnam commented 5 years ago

No @THUHJ