yolo darknet training pipeline:

• annotate large images
• copy images to new location
• rescale images
• reformat annotations
• create data files

bounding box formats

rectlabel

• notation (pixel): xmin, ymin, xmax, ymax;
• origin: top left

darknet

• notation (relative): class-index, xcentre, ycentre, width, height
• origin: top left

how-to run with darknet

1. clone and make darknet in dir over set-solver (https://pjreddie.com/darknet/install/)

2. copy pretrained weights from yolo website (guide: https://pjreddie.com/darknet/yolov2/, weights: https://pjreddie.com/darknet/imagenet/#darknet19_448)

3. run ./darknet detector train ../set-solver/yolo/set.data cfg/yolov2.cfg darknet19_448.conv.23

4. wait untill atleast iteration 100 before quitting to save the progress

5. test with ./darknet detect cfg/yolov2.cfg backup/

this, its the plan

• configure cloud server for training
• mvp object detector with darknet/yolo
• try implement cnn with mobile net
• make the app
• win set-game