tensorflow QAT model like:
【model 1】
[data] [weights]
| |
[FakeQuantWithMinMaxVars] [FakeQuantWithMinMaxVars]
\ /
[ conv2d ]

I hope it to be onnx QAT model like this after convert:
【model 2】
[data] [weights]
| |
[QuantizeLinear] [QuantizeLinear]
| |
[DequantizeLinear] [DequantizeLinear]
\ /
[ conv2d ]

But it turned to be like:
【model 3】
[data]
|
[QuantizeLinear]
|
[DequantizeLinear]
\
[ conv2d ]

Seems like weight has const folded .
I noticed that there is an arg in tf2onnx.convert :

--fold_const Deprecated. Constant folding is always enabled.

Does this have anything to do with it?

please help, thanks.