onnx result and torch result don't match

[MengWangTHU]

Hi, Carkusl:
thanks a lot for your work.
I use the same tools and the same config to turn torch model to onnx model; I get pfe_onnx and rpn_onnx. Before simplify the onnx model, I use onnxruntime to test the onnx result to validation if the reuslt consistent with the torch model. But I got the negative results.
Wheather the pfe or the rpn, the onnx inferenc results are very different with the torch results.
I have not do the simplify. dose the simiplification work for the big difference of the model-inference-result?

Do you have any idea about the problem.

[CarkusL]

I think the simplification doesn't affect the results.
can you share your onnx models?

[MengWangTHU]

I think the simplification doesn't affect the results.
can you share your onnx models?

yeah, it's a pleasure to share my onnx models, incuding pfe and rpn. But how can i send them to you?

[MengWangTHU]

I think the simplification doesn't affect the results.
can you share your onnx models?

yeah, it's a pleasure to share my onnx models, incuding pfe and rpn. But how can i send them to you?

Your email address?

[CarkusL]

I think the simplification doesn't affect the results.
can you share your onnx models?

yeah, it's a pleasure to share my onnx models, incuding pfe and rpn. But how can i send them to you?

Your email address?

you can upload your models to google drive