Built for object detection models, other workloads may not be ideal. Mainly for visualizing non-TensorFlow models on TensorBoard, as well as other TensorFlow-specific tools.
This tool takes a Caffe prototxt/Onnx model and generates a TensorFlow GraphDef which is serialized and written to a protobuf (converted_model.pb by default). Not configured for inference use.
Common usage involves running the following commands:
- $ python3 caffe2tf.py -m path/to/deploy.prototxt or
- $ python3 onnx2tf.py -m path/to/model.onnx
- converted_model.pb (default name) is generated in the same directory
- Caffe models only require a .prototxt file. Caffemodel files are not required.
- Onnx models only require a .onnx file.
- -m : This is a required argument reflecting the path to your Caffe prototxt/Onnx model file
- -o : This is an optional argument to set the output TensorFlow protobuf's name
- caffe2tf.py
- onnx2tf.py