Stuff to prepare for training object detection (Set burnin to 0 for training rate to start high)
-BBoxLabelTool for labeling images: https://github.com/puzzledqs/BBox-Label-Tool
-training images with labels: https://timebutt.github.io/content/other/NFPA_dataset.zip
-clone https://github.com/AlexeyAB/darknet
-make (GPU=1) (CUDNN=1) (for GPU and CUDA Support)
-mkdir data/obj
-copy yolo-obj.cfg to cfg folder (adjust batch, subdivisions, classes, filters) (It's the yolo v3 config with changes)
-copy obj.names to cfg (adjust object names)
-copy obj.data to cfg (adjust class number)
-put .jpg images in directory data/obj
-add labels to data/obj/
-(convert labels? https://github.com/Guanghan/darknet/blob/master/scripts/convert.py)
-run process.py in obj/data folder to create train.txt and test.txt
wget https://pjreddie.com/media/files/darknet53.conv.74 (to darknet/)
-start training: ./darknet detector train cfg/obj.data cfg/yolo-obj.cfg darknet53.conv.74
-result is located in backup?
shell:
git clone https://github.com/phixxx5/howToTrainYolo
wget https://timebutt.github.io/content/other/NFPA_dataset.zip
unzip NFPA_dataset.zip
git clone https://github.com/AlexeyAB/darknet
cd darknet
make
mkdir data/obj
cd ..
cp howToTrainYolo/yolo-obj.cfg howToTrainYolo/obj.names howToTrainYolo/obj.data darknet/cfg/
cp NFPA\ dataset/* darknet/data/obj
cp howToTrainYolo/process.py darknet/data/obj
cd darknet
python2.7 data/obj/process.py
wget https://pjreddie.com/media/files/darknet53.conv.74
./darknet detector train cfg/obj.data cfg/yolo-obj.cfg darknet53.conv.74