Awesome
CAFFE for YOLO9000
Reference
YOLO9000: Better, Faster, Stronger
Usage
Caffe
vim Makefile.config
make clean
make all -j8
make pycaffe
(Be careful: caffe&caffe2 PYTHONPATH conflict)
Data preparation
Like SSD data setting.
cd data/VOC0712
vim create_data.sh
./create_data.sh
Train
cd examples/yolo/darknet_voc
mkdir models
# edit according yourself
vim gnet_region_train_darknet448_voc.prototxt
vim gnet_region_test_darkent448_voc.prototxt
vim gnet_region_solver_darkent448_voc.prototxt
# download pretrain_model
https://pan.baidu.com/s/1c71EB-6A1xQb2ImOISZiHA password: 9u5v
# change related path in script train.sh
vim train_darknet448.sh
./train_darknet448.sh
Eval mAP
cd demo
vim demo_yolo.py
python demo_yolo.py
Test a image
vim test_darknet448.sh
./test_darknet448.sh