Hi, I'm changing Deep-Learning Framework from Caffe to PyTorch. So, all the models in Caffe is porting into PyTorch but several problems happen. Specially in my own detection models using SSD, 'PriorBox' and 'DetectionOutput' layers are not directly supported by PyTorch but supported in ONNX operators according to 'Supported Framework Layers' in OpenVINO's document.
If so, how do I make Model Optimizer generate IR using those operators ? Is there any way except for making PyTorch's custom operator for ONNX myself ?
Link Copied
Hi Byung Hun,
Thanks for reaching out to us.
Support for PriorBox and DetectionOutput layers as ONNX custom operations has been implemented recently, as per https://github.com/openvinotoolkit/openvino/pull/2064
I would suggest you try convert your PyTorch model with custom operations to ONNX format, and subsequently to IR using Model Optimizer.
Regards,
Munesh
Hi Byung Hun,
This thread will no longer be monitored since we have provided a solution. If you need any additional information from Intel, please submit a new question.
Regards,
Munesh
For more complete information about compiler optimizations, see our Optimization Notice.