ch_ppstructure_mobile_v2.0_SLANet表格识别模型转rknn模型报错

[cty-ai]

Please fill in the information below so that we can solve the problem quickly, Thanks !

Describe the bug
模型：ch_ppstructure_mobile_v2.0_SLANet
Informations (please complete the following information):

• Inference engine for deployment:
• Why convert to onnx：硬件是rk3588
• Paddle2ONNX Version:1.1.0
• Email/Wechat/Phone:[email protected]

Screenshots

Additional context
rknn_toolkit2-1.5.3b21+11d8562a-cp38-cp38-linux_x86_64.whl
rknn_toolkit2-1.6.0+81f21f4d-cp38-cp38-linux_x86_64.whl
rknn_toolkit2-1.6.1b16+e3c0774a-cp38-cp38-linux_x86_64.whl
rknn_toolkit2-1.6.2b3+eeda9fd5-cp38-cp38-linux_x86_64.whl
这几个toolkit都试过报一样的错误，转onnx没报错，再转rknn报错如截图

[Zheng-Bicheng]

这个已经很明显了呀，你需要固定一下Shape再进行转换，RKNN只支持固定的shape,这和Paddle2ONNX无关。