ch_PP-OCRv4_det 原始超轻量模型的训练 #12973
Replies: 10 comments
-
PP-OCRv4的骨干网络带了重参数化(rep)模块,你这边展示的这个是rep前的模型,动态图的pdparams会比较大,可以导出成静态图模型,就会比较小了,也就是真实使用的时候是比较小的。 |
Beta Was this translation helpful? Give feedback.
-
你好,请问如何将动态图转成静态图呢? |
Beta Was this translation helpful? Give feedback.
-
导出成静态图模型的化存储差不多下降4倍左右,你训练的字典有多大呢? |
Beta Was this translation helpful? Give feedback.
-
@cuicheng01 训练的是det模型,det应该没有字典? |
Beta Was this translation helpful? Give feedback.
-
@cuicheng01 导出静态图是tools/export_model.py这个脚本吗 |
Beta Was this translation helpful? Give feedback.
-
已导出,得到三个inference模型,分别是Teacher、Student和Student2,两个student模型确实是4.8M |
Beta Was this translation helpful? Give feedback.
-
你好,你在训练的时候使用 ch_PP-OCRv4_det_cml.yml 这个配置文件不会报错吗? |
Beta Was this translation helpful? Give feedback.
-
按照文档设置好数据路径,一般不会出错 |
Beta Was this translation helpful? Give feedback.
-
我这边报 Backbone不能使用PPLCNetNew网络,请问你的paddleocr版本是多少? |
Beta Was this translation helpful? Give feedback.
-
将 |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
用自己的数据集从头开始训练ch_PP-OCRv4_det原始超轻量模型,根据如下表格,设置配置文件后,开始训练
python tools/train.py -c configs/det/ch_PPpOCRv4/ch_PP-OCRv4_det_cml.yml
训练保存的模型如下,不是超轻量模型,请问如何设置训练超轻量模型?
Beta Was this translation helpful? Give feedback.
All reactions