Problem on text detection finetuning - English #13168
Answered
by
tzktok
kevinzezel
asked this question in
Q&A
-
问题描述 / Problem DescriptionHi, Which .yml file should I use to perform text detection finetuning in English? And which pretrained model should I start to do the finetuning in English? I tried using the template ch_PP-OCRv3_det_student.yml and pre trained model en_PP-OCRv3_det_distill_train/student.pdparams. However, the loss error never drops with each iteration, it is always fixed at 1. 运行环境 / Runtime Environment
复现代码 / Reproduction Code
完整报错 / Complete Error Message |
Beta Was this translation helpful? Give feedback.
Answered by
tzktok
Jun 24, 2024
Replies: 3 comments 4 replies
-
|
how to u annotate the datset..could u explain.. thx in advance @kevinzezel |
Beta Was this translation helpful? Give feedback.
0 replies
-
|
|
Beta Was this translation helpful? Give feedback.
0 replies
-
|
@tzktok @GreatV Thanks for the quick response. Follow my files Sample of Train.txt: Sample of Val.txt: Sample of img dataset: Pretrained model: ch_PP-OCRv3_det_student.yml |
Beta Was this translation helpful? Give feedback.
4 replies
-
|
@kevinzezel how u annotate ur own dataset..using ppocrlabel ? |
Beta Was this translation helpful? Give feedback.
Answer selected by
kevinzezel
-
|
I'm using my own script to generate the training base based on the documentation I read on github. But now that you asked, I went to take a look at ppocrlabel and this example: And I saw that I was making the wrong coordinate by "Points". I made the correction and now it works! Thank you very much! Regards, |
Beta Was this translation helpful? Give feedback.
-
|
Hi Kevin I am also training a detection model, and I would like to know what “wrong coordinate by ‘Points’” means.
in official docs? thanks! |
Beta Was this translation helpful? Give feedback.
-
Hello Kevin, Can you please emphasize on "And I saw that I was making the wrong coordinate by "Points" ". I am going thro the same difficulty of not having any text detected and I am using the same model: en_PP-OCRv3_det_distill_train/student.pdparams with ch_PP-OCRv3_det_student.yml as config file. I have done data annotation same as you did. I performed this on a sample of 10 image datasets and 7 for validation. Currently this fine-tuned model does text detection poor than baseline paddle model so I wanted to enquire on this. Can you please help me? |
Beta Was this translation helpful? Give feedback.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
@kevinzezel how u annotate ur own dataset..using ppocrlabel ?