search

PaddlePaddle / PaddleOCR

Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)
https://paddlepaddle.github.io/PaddleOCR/
Apache License 2.0
44.1k stars 7.81k forks source link

怎么知道yml配置文件对应使用什么模型 #8195

Closed YFforever2022 closed 1 year ago

YFforever2022 commented 2 years ago

请问怎么知道yml配置文件对应使用什么模型

gavinfaimdata commented 2 years ago

look at the title of the config file

YFforever2022 commented 2 years ago

z1 比如图片上这样，我是根据这样来选择对应的预训练模型的，这样应该没错吧

ch_PP-OCRv3_det_cml.yml 使用 ch_PP-OCRv3_det_distill_train，训练ch_PP-OCRv3_det训练2600次预测出来的precision:0.3717948717948718

det_mv3_db.yml 使用 MobileNetV3_large_x0_5_pretrained.pdparams，训练5800次预测出来的precision:0

这让我产生了怀疑，也许就是预训练权重不对应导致的

请问文件标题文字相似度高权重文件的就是吗

gavinfaimdata commented 2 years ago

If the weights from the pretrained model were off you'd still get > 0 precision, I think. I have only used the config files in PaddleOCR/configs/det, I haven't used the ones shown in this table.