Model accuracy lower with TRT7

Hi

We got lower accuracy with TRT7 than with TRT6 on Nano as using LFFD.
Is there any to adjust it?

Thank you,

Hi,

What kind of model do you use? Ex. pb, onnx, uff, …

Usually, you should get the identical accuracy between TensorRT-6 and TensorRT-7.
Could you help to verify which layer cause the degradation first?

Thanks.

Hi AastaLLL,

Thanks for your prompt support.

The model uses onnx.
Would you address more how to do the verify of layer?
Due to we did not have experience on it?

Thank you,

Hi,

To give a further suggestion, could you also share the onnx, input, and the corresponding source with us?

Thanks.

Hi AastaLLL,

The input size is 1280x720.
The source is
GitHub - YonghaoHe/LFFD-A-Light-and-Fast-Face-Detector-for-Edge-Devices: A light and fast one class detection framework for edge devices. We provide face detector, head detector, pedestrian detector, vehicle detector......

Thank you,

Working on this through the private message directly.

Hi,

The root cause of this issue is located.
Please check the private message for the detailed workaround.

Thanks.

Hi AastaLLL,

Thank you so much for your great support.
The solution works.

Thanks