Exporting PARSEQ Recognition Model to ONNX

[leduy-it]

请将下面信息填写完整，便于我们快速解决问题，谢谢！

问题描述
请在此处详细的描述报错信息

[Paddle2ONNX] Start to parse PaddlePaddle model...
[Paddle2ONNX] Model file path: ./paddle_inference/weights/paddle_inference/parseq_171224/inference.pdmodel
[Paddle2ONNX] Parameters file path: ./paddle_inference/weights/paddle_inference/parseq_171224/inference.pdiparams
[Paddle2ONNX] Start to parsing Paddle model...
[Paddle2ONNX] DenseTensorArray is not supported.
[Paddle2ONNX] Oops, there are some operators not supported yet, including lod_array_length,memcpy,tensor_array_to_tensor,while,write_to_array,
[ERROR] Due to the unsupported operators, the conversion is aborted.


--------------------------------------
C++ Traceback (most recent call last):
--------------------------------------
0   paddle2onnx::Export(char const*, char const*, char**, int*, int, bool, bool, bool, bool, bool, paddle2onnx::CustomOp*, int, char const*, char**, int*, char const*, bool*, bool, char**, int)

----------------------
Error Message Summary:
----------------------
FatalError: `Process abort signal` is detected by the operating system.
  [TimeInfo: *** Aborted at 1735310555 (unix time) try "date -d @1735310555" if you are using GNU date ***]
  [SignalInfo: *** SIGABRT (@0xbb) received by PID 187 (TID 0x727a43df9740) from PID 187 ***]

Aborted (core dumped)

更多信息 :

• 用于部署的推理引擎: 待更新

• 为什么需要转换为ONNX格式：Deploying in ONNX Runtime

• Paddle2ONNX版本:
  Name: paddle2onnx
  Version: 1.3.1

• 你的联系方式(Email/Wechat/Phone): [email protected]

报错截图

其他信息

• I am working on integrating PARSEQ OCR recognition with PaddleOCR.

I came across Issue #12, which discusses ONNX export handling in PyTorch. I'm exploring ONNX export for PARSEQ but want to retain AR mode functionality and refinement iterations without compromise.

Would appreciate guidance or suggestions to achieve this.

Thank you!