This guide will show you how to convert your models to Anakin models.
Before using Anakin, you must convert your models to Anakin ones. If you don't, Anakin won't work properly.
- Python 2.7+
- Protobuf 3.1+ (Make sure that Pip Protobuf is consistent with the system Protobuf version.)
- PaddlePaddle 0.12.0+ (Fluid mode)
- flask, bson, matplotlib, scikit-image
- tkinter
git clone https://xxxxxxxxxConfigure your config.yaml file. Find example config.yaml file in the converter source directory. The example below explains how to configure your config.yaml file.
OPTIONS:
Framework: CAFFE # select a target dl-framework you want parsing
SavePath: ./output
ResultName: googlenet # the name you want when saving the parsed model
Config:
LaunchBoard: ON # should be on if you want to launch graph board
Server:
ip: 0.0.0.0
port: 8888
OptimizedGraph: # only enable(set enable(ON) and path) when you have optimized graph model.
enable: ON
path: /path/to/anakin_optimized_anakin_model/googlenet.anakin.bin.saved
LOGGER:
LogToPath: ./log/ # the path where log
WithColor: ON # colorful log message
TARGET:
CAFFE:
# path to proto files
ProtoPaths:
- /path/to/caffe/src/caffe/proto/caffe.proto
PrototxtPath: /path/to/your/googlenet.prototxt
ModelPath: /path/to/your/googlenet.caffemodel
FLUID:
# path to proto files
Debug: NULL
ModelPath: /path/to/model
# ...After finishing configuration , you just need to call python script python converter.py to complete transfromation.
Anakin external converter will be launched on site http://0.0.0.0:8888 (configurable). Then open you browser and search http://0.0.0.0:8888, amazing things will happen!
if you set ip to 0.0.0.0 in remote server, you need to open local browser and search the server real ip:port, not the 0.0.0.0.