convertToEngineFromUff with int8 mode when Meeting "uniformly zero, network calibration failed"

When I did convertToEngineFromUff with int8 mode, I met following error. How can I fix it? My model can run successfully in fp32 mode.

ERROR: Tensor FeatureExtractor/InceptionV2/Mixed_4a/Branch_1/Conv2d_0a_1x1/Relu6/relu2 is uniformly zero; network calibration failed.


Please generate a calibration table for your model.

When creating networks in 8-bit integer precision, it uses a process called calibration to determine the dynamic range of intermediate activations, and hence the appropriate scaling factors for quantization.

Here are more information for your reference: