YoloCut

Tool for cutting videos from frames which don't present desired objects. The intention was to automatically preprocess skate videos to cut out scenes without person and skateboard. I based my implementation on the YoloV3 examples from https://github.com/spmallick/learnopencv/tree/master/ObjectDetection-YOLO.

Example:

input https://youtu.be/tob3PN2lnsc

output https://www.youtube.com/watch?v=UnfoIU_GHHA

cut out scenes https://www.youtube.com/watch?v=eP5Uh_KXV5A

usage:

docker build . -t yolo_cut
docker run -v ~/Downloads:/input/ -e INPUT_VIDEO=/input/skate_vine.mp4 -e OUTPUT_VIDEO=/input/skate_vine_processed.mp4 -e KEYWORDS="skateboard person" yolo_cut

TODO:

• Try yolo9000
• Use ffmpeg to cut video
• Move config to json
• Read args
• Train model on my own dataset
• Detect if person is on skateboard
• Speed it up