Skip to content

P
PaddleOCR

PaddlePaddle / PaddleOCR
大约 1 年 前同步成功

通知 1528
Star 32962
Fork 6643
P
PaddleOCR

体验新版 GitCode，发现更多精彩内容 >>

提交 b6cd6047 编写于 作者: L LDOUBLEV
浏览文件
操作

cml config

上级 293e85bc
隐藏空白更改
内联 并排
Showing with 53 addition and 50 deletion
configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml
浏览文件 @ b6cd6047
...@@ -19,57 +19,60 @@ Global: ...@@ -19,57 +19,60 @@ Global:
distributed: true distributed: true
Architecture: Architecture:
name: DistillationModel
algorithm: Distillation
model_type: det model_type: det
algorithm: DB Models:
NetWorks: Student:
Student: model_type: det
model_type: det algorithm: DB
algorithm: DB Transform: null
Transform: null Backbone:
Backbone: name: MobileNetV3
name: MobileNetV3 scale: 0.5
scale: 0.5 model_name: large
model_name: large disable_se: true
disable_se: true Neck:
Neck: name: CAFPN
name: CAFPN out_channels: 96
out_channels: 96 shortcut: True
shortcut: True Head:
Head: name: DBHead
name: DBHead k: 50
k: 50 Student2:
Student2: model_type: det
model_type: det algorithm: DB
algorithm: DB Transform: null
Transform: null Backbone:
Backbone: name: MobileNetV3
name: MobileNetV3 scale: 0.5
scale: 0.5 model_name: large
model_name: large disable_se: true
disable_se: true Neck:
Neck: name: CAFPN
name: CAFPN out_channels: 96
out_channels: 96 shortcut: True
shortcut: True Head:
Head: name: DBHead
name: DBHead k: 50
k: 50 Teacher:
Teacher: pretrained: ./inference/ch_ppocr_resnet50/student
pretrained: ./inference/ch_ppocr_resnet50/student freeze_params: true
model_type: det return_all_feats: false
algorithm: DB model_type: det
Backbone: algorithm: DB
name: ResNet Backbone:
in_channels: 3 name: ResNet
layers: 50 in_channels: 3
Neck: layers: 50
name: FEPAN Neck:
out_channels: 256 name: FEPAN
Head: out_channels: 256
name: DBHead Head:
kernel_list: [7,2,2] name: DBHead
k: 50 kernel_list: [7,2,2]
k: 50
Loss: Loss:
name: CombinedLoss name: CombinedLoss
loss_config_list: loss_config_list:
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或