Here, I have some experiments in order to make more accurate classifier. Triplet Loss, Batch Triplet, SVM are used for experiments. softmax features + SVM was the best in my result as a record of 91% top1 accuracy.

Kaggle's State Farm Distracted Driver Detection Dataset is used for evaluating the performance of Image Classification, you should modify sampledata.py to fit your dataset.

Rebuild your caffe directory:

cd $CAFFEROOT$
cp Makefile.configexample Makefile.config

Remember to uncomment the line to makesure your python layers could be found:

WITH_PYTHON_LAYER := 1

Then build caffe and pycaffe:

make all -j8 & make pycaffe

1. Modify sampledata.py, config.py and train.py to fit your dataset and working environment.

2. Pre-train your model with softmax loss.

3. Finetune triplet model based on your pre-trained model.

4. Learn to adjust parameters.

softmax

triplet loss

batch triplet loss

softmax+svm

This project is based on hizhangp's job