this repo provides

• python scripts
• .pt model

• car detection model: link
• large vehicle head and body detection model: not uploaded

• use requirement.txt ( only way now )
• use shell.nix( without ultralytics package, can't do models things )

clip-video.py

• extract a specific video clip
• Usage: python3 clip-video.py <video path> <start time> <end time> <output directory>

get-frame.py

• get a specific frame picture from video(or process a folder of video)
• Usage: python3 get-frame.py <video_path>/<video_folder_path> <time_str> <output_path>
• time_str can be exactly '3:42' or vague '1/2'

predict-image.py

• predict objects in image
• Usage: python3 predict-picture.py <image-path> <model-path>

track.py

• do tracking
• Usage: python3 track.py <model path> <source video>

train.py

• train models
• Usage: python3 train.py <model path> <yaml path> <epoches>

train-resume.py

• resume train models
• Usage: python3 train-resume.py <model path>

extract-car-image.py

• extract all single car images from a whole image
• Usage: python3 extract-car-image.py <model path> <image path> <output dir path> <scale factor>

everything in utils is for specific things, not for general