Fine-Tuning of Paddle OCR: Issues with text recognition #13875
-
Hi guys, I am new to this space and I am currently exploring on fine-tuning of PaddleOCR. I have read about the implementation procedures already on: https://github.com/PaddlePaddle/PaddleOCR/blob/main/doc/doc_en/finetune_en.md My query is very general though.
Context: I used en_PP-OCRv4_rec text recognition: https://github.com/PaddlePaddle/PaddleOCR/blob/main/doc/doc_ch/models_list.md and used en_PP-OCRv3_rec.yml as modified yaml file! Thanks in advance for guidance! @GreatV @WenmuZhou @LDOUBLEV @MissPenguin @tink2123 ..... can someone please help me? |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments 5 replies
-
### Config file: Global:
Optimizer: Architecture: Loss: PostProcess: Metric: Train: ### Train Log: [2024/09/11 16:13:22] ppocr INFO: cur metric, acc: 0.0, norm_edit_dis: 7.692301774442356e-07, fps: 16.36477335822396 |
Beta Was this translation helpful? Give feedback.
-
This warning is because you use the v3 configuration file to fine-tune the v4 model, resulting in the inability to correctly match the parameters when loading the pre-trained model. You need to use the v4 configuration file. |
Beta Was this translation helpful? Give feedback.
-
@VishyAnand28 For me, it seems like the parameters are not being loaded successfully. The accuracy of the first epoch is 0. Was it the same for you? |
Beta Was this translation helpful? Give feedback.
@UserWangZz thanks a ton for the swift response and also for clarifying my queries. Have a great day ahead!