Skip to content

bugfix:solve v5 cls model input size issue.#8

Open
OrchesAdam wants to merge 1 commit into
RapidAI:mainfrom
OrchesAdam:main
Open

bugfix:solve v5 cls model input size issue.#8
OrchesAdam wants to merge 1 commit into
RapidAI:mainfrom
OrchesAdam:main

Conversation

@OrchesAdam
Copy link
Copy Markdown

AngleNet 中分类模型的输入尺寸被硬编码为 192×48,只能兼容 PP-OCRv2/v4 的 cls 模型。PP-OCRv5 换用了新的 PP-LCNet 架构,输入尺寸变为
160×80,导致 ONNX 推理时报 InvalidArgument 错误。

根因

AngleNet.cs 用编译期常量写死了 resize 尺寸:

private const int angleDstWidth = 192;
private const int angleDstHeight = 48;

而 PP-OCRv5 的 cls 模型(ch_PP-LCNet_x*_textline_ori_cls_*.onnx)期望 160×80 输入,导致:

Got: 48 Expected: 80
Got: 192 Expected: 160

修复

改为在初始化时从 ONNX 模型 metadata 自动读取输入尺寸:

  • 在 InitModel 中从 InputMetadata 读取维度:
var dims = angleNet.InputMetadata.First().Value.Dimensions;
_dstHeight = dims[2];  // v2=48, v5=80
_dstWidth = dims[3];   // v2=192, v5=160

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant