- Ask a related questionWhat is a related question?A related question is a question created from another question. When the related question is created, it will be automatically linked to the original question.
Tool/software: TI C/C++ Compiler
Hello,
I want to evaluate my INT8 detection model after QAT. I refer to the codes in pytorch-mmdetection, and find that "model = XMMDetQuantTestModule(model, dummy_input)" is used for evaluation. So I try to do the inference like
self.model = xnn.quantize.QuantTestModule(self.model, dummy_input=dummy_input, forward_analyze_mothod='forward_dummy') # wrap the model
self.model.module = load_model(self.model.module, opt.load_model) #load the model parameters saved by QAT
self.model = self.model.to(opt.device)
self.model.eval()
The detection results seems to be good, but I am not sure if the inference is done in INT8 like that on TDA4. And when i import the ONNX model and run it on TDA4, the performance decreases a lot.
So do you have any suggestions about this?
Thanks.
Best regards,
Zorro.
