- symbol转prototxt: json2prototxt.py, 转出后手动把不支持的地方接上
- 权重转换:mxnet2caffe.py
- 如果想两步合一:mxnet2caffe_all_in_one.py,如果有不支持的层就不能2步合一了
- 如果需要调整输入顺序:caffe_swap_channel_toolkit.py
- 如果需要inplace操作:inplace_conv_block.py, 注意转换后可能需要手动微调
- insert_mean_scale.py, 将均值和方差用Scale层的形式写入模型文件
- mxnet2caffe_all_in_one_experimental.py: 与步骤三功能一致,区别是自动跳过不支持的层
English Guidelines:
- symbol file convert to prototxt: json2prototxt.py, if there are some layers skipped, modify the prototxt file mannualy
- convert params: mxnet2caffe.py
- 1,2 steps in one: mxnet2caffe_all_in_one.py, however, if there are some layers skipped, we cannot use it.
- swap input channel(bgr->rgb or vice versa): caffe_swap_channel_toolkit.py
- inplace op: inplace_conv_block.py
- use Scale layer to process data mean and std: insert_mean_scale.py
- mxnet2caffe_all_in_one_experimental.py: same with step 3 but skip the unsupported layers automaticaly