Convert ONNX to Caffe

This tool converts ONNX models to Caffe format. Only use for inference.

Dependencies

• Caffe 1.0 (with Python support)
• ONNX 1.4.0
• protobuf 3.7.1

How to use

To convert an ONNX model to Caffe:

python convertCaffe.py --onnx ./model/MobileNetV2.onnx --prototxt ./model/MobileNetV2.prototxt --caffemodel ./model/MobileNetV2.caffemodel

Setup with Docker

Build a Docker image with a minimal Caffe 1.0, ONNX 1.4.0 and protobuf 3.7.1 installation, using the Dockerfile inside this repo:

docker build --network=host -t onnx-caffe_18.04 .

Then create a Docker container based on this image and open a terminal inside:

docker run --network=host -e PYTHONPATH=/workspace/onnx2caffe -v $PWD:/workspace -it onnx-caffe_18.04 /bin/bash
cd /workspace
python3 convertCaffe.py [with arguments from above]

Currently supported operations

• Conv
• ConvTranspose
• BatchNormalization
• MaxPool
• AveragePool
• Leaky ReLU
• ReLU
• Sigmoid
• Dropout
• Gemm (InnerProduct only)
• Add
• Mul
• Reshape
• Upsample
• Concat
• Flatten