Mercor Challenge Repository

Type of Meta data I Used Are

1. Visual Hotspot/Deadzone Detection

• Scene Edit Detection - To complete this task i used PySceneDetect Adaptive or content Algorithm by minor changing in the internal structure of the pyscenedetect i did try to use manual diffrent algorithm but they are not accurate so i went with this Pyscenedetect

• Focus Detection - To complete this task i used Fast Fourier Transform(FFT) Algorthim which use Laplacian method to detect the blue on the thresold of 10 you can change thresold acordingaly

2. Person Detection(per frame) - To work on this i used Yolov8 Which is faster than any other ai model on the GPU

If i consider only visual hotspot detection or only Focus detection its very fast and work on Real Time Detection but using both the Detection Method Its make it somewhat Slower

Installation requirement - python 3.8.0+

1. Clone the Repository

git clone <git repo link>

2. Then install all the dependencies

pip install -r requirements.txt

3. Then Import Engine class in your Python File as like as app.py file and add list of video path or single path in the engine class and it will generate

from  Engine import Engine


lists=['video.mp4','video2.mp4']

Engine(lists,yolo=True)

You can deselect yolo model as it take gpu to run by make the parameter yolo = False

Engine(lists,yolo=False)

It will generate Json File of Metadata of particual file with the name of the path of the file