[Build] TypeInferenceError when quantize an onnx model with custom operator #23191
Description
Describe the issue
I use onnxruntime python api to quantize a model with custom operator.
@onnx_op was used in my code to register the custom op:
@onnx_op(op_type='LocatePostNms',
domain='ai.onnx.contrib',
inputs=[PyCustomOpDef.dt_float, PyCustomOpDef.dt_int32, PyCustomOpDef.dt_float, PyCustomOpDef.dt_float,
PyCustomOpDef.dt_int32, PyCustomOpDef.dt_float],
outputs=[PyCustomOpDef.dt_float, PyCustomOpDef.dt_int32],
attrs={"classNumber": PyCustomOpDef.dt_int32,
"iou_thresh": PyCustomOpDef.dt_float,
"scale": PyCustomOpDef.dt_double,
"side": PyCustomOpDef.dt_int32,
"topk": PyCustomOpDef.dt_int32,
"type": PyCustomOpDef.dt_int32}
)
def locatepostnms_compute(x, **kwargs):
padded_box = np.ones((1, 24000, 7))
result_cnt = np.ones(1)
return padded_box, result_cnt
and set sess_options.register_custom_ops_library(get_library_path()), but error happened when running onnxruntime. quantization. quantize. quantize_static interface, report:
onnx.onnx_cpp2py_export.shape_inference.InferenceError: [TypeInferenceError] Cannot infer type and shape for node name 453. No opset import for domain ai.onnx.contrib optype LocatePostNms
where 453 is the name of custom op, how do i infer type and shape for this type of onnx model to finish the quantization?
Sorry that i can't provide the model
Urgency
No response
Target platform
windows 11 PRO/ ubuntu 22.04
Build script
python
Error / output
onnx.onnx_cpp2py_export.shape_inference.InferenceError: [TypeInferenceError] Cannot infer type and shape for node name 453. No opset import for domain ai.onnx.contrib optype LocatePostNms
Visual Studio Version
No response
GCC / Compiler Version
No response