Convert ONNX to Caffe

This tool is modified from onnx2caffe by MTlab.

We added some OPs to support one-stage mmdetection models.

Dependencies

• pycaffe (with builtin Upsample and Permute layers)
• onnx

How to use

To convert onnx model to caffe:

python convertCaffe.py ./model/mmdet.onnx ./model/a.prototxt ./model/a.caffemodel

Current support operation

• Conv
• ConvTranspose
• BatchNormalization
• MaxPool
• AveragePool
• Relu
• Sigmoid
• Dropout
• Gemm (InnerProduct only)
• Add
• Mul
• Reshape
• Upsample
• Concat
• Flatten
• Resize
• Permute
• Scale