Один из авторов - Князев Александр АСУ2-19-1м
Folder
dataset_generation/
All programs to annotate and generate data are placed in this folder.
Folder
yolo/
To train YOLO you should generate files with annotation. Use this program:
voc_to_YOLOv3.py
If you have one class of object you can use this pgrogam to train yolo model:
train.py
To train model whith more than one class use this program:
train_combine_2.py
Folder
dlib/
All programs to train dlib shape predictor for kepoints detection (required images annotated keypoints).
Folder
evaluation/
All programs to evaluate yolo object detection models using Intersection ovaer Union and dlib shape predictor.
Folder
final_system/
All programs to run final system which detects object (luminaire), finds it's keypoints and draws the 3D bounding box. To run final system, use:
python realtime_detect_with_kpd_pos.py --shape-predictor kplum_gen_ninth_comb_test.dat
but it requires trained YOLO object detection model (.h5) in folder logs/ and trained dlib shape predictor (.dat)
In folder
intel_realsense_camera/
you can find some programs which allows to capture data from intel realsense camera.