Image Classification Example Using the C Predict API
This is a simple predictor which shows how to use the MXNet C Predict API for image classification with a pre-trained ImageNet model in a single thread and multiple threads.
- OpenCV for image processing:
USE_OPENCVis set to true by default when building from source
How to Use this Example
Download the Model Artifacts
- You will need the model artifacts for the Inception ImageNet model. You can download these from http://data.mxnet.io/mxnet/models/imagenet/inception-bn/
- Place them into a
model/Inception/subfolder, or if not, you will need to edit the source file and update the paths in the Build step.
- If using a different location for the model artifacts, edit
image-classification-predict.ccfile, and change the following lines to your artifacts' paths:
// Models path for your model, you have to modify it std::string json_file = "model/Inception/Inception-BN-symbol.json"; std::string param_file = "model/Inception/Inception-BN-0126.params"; std::string synset_file = "model/Inception/synset.txt"; std::string nd_file = "model/Inception/mean_224.nd";
- You may also want to change the image size and channels:
// Image size and channels int width = 224; int height = 224; int channels = 3;
- Simply just use our Makefile to build:
Run the example by passing it an image that you want to classify. If you don't have one handy, run the following to get one:
Then run the
image-classification-predict program, passing the image as the first argument and the number of threads as the second parameter.
./image-classification-predict 1920px-Honeycrisp.jpg 1
- If you don't run it in the MXNet root path, you may need to copy the
pertusa (for Makefile and image reading check)
caprice-j (for reading function)
sofiawu (for sample model)
piiswrong and tqchen (for useful coding suggestions)