PaddleOCR/configs
sylarwcy 68b384292b
v4 det cml configs (#11258)
* fixed several bugs

1. 修复了找不到模型的问题,原因是PPLCNetNew不在可选模型名称的列表内,将PPLCNetNew改为PPLCNetV3;
2. 修复了db_fpn.py解析in_channels的报错问题,db_fpn.py中按序列解析in_channels,而lcnetv3.py中det为false时输出是数值,为true才返回序列。在两个student模型的backbone中添加"det: true";
3. 减少CPU占用,将cal_metric_during_train: true 改为 false;
4. 修复了训练过程中eval时的显存溢出问题。通过限制过大的测试数据可解决该问题,具体调整是,在eval→DetResizeForTest的配置中增加"limit_side_len: 960,limit_type: max"。

* 恢复cal_metric_during_train的设置
2023-11-17 20:38:25 +08:00
..
cls
det v4 det cml configs (#11258) 2023-11-17 20:38:25 +08:00
e2e
kie fix performance (#9772) 2023-04-20 09:53:31 +08:00
rec add cppd u14m train model and doc (#11052) 2023-10-11 17:15:01 +08:00
sr add and update sr model 2022-10-17 16:47:28 +08:00
table fix performance (#9772) 2023-04-20 09:53:31 +08:00