app.py

import gradio as gr
import os
import cv2
import face_recognition
from fastai.vision.all import load_learner
import time

model = load_learner("gaze-recognizer-v1.pkl")

def video_processing(video):
    start_time = time.time()
    # Loop through the frames of the video
    video_capture = cv2.VideoCapture(video)
    on_camera = 0
    off_camera = 0
    total = 0
    while True:
        # Read a single frame from the video
        for i in range(24*30):
            ret, frame = video_capture.read()
            if not ret:
                break

        # If there are no more frames, break out of the loop
        if not ret:
            break
        
        # Convert the frame to RGB color (face_recognition uses RGB)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        

        # Find all the faces in the frame using a pre-trained convolutional neural network.
        face_locations = face_recognition.face_locations(gray)
        #face_locations = face_recognition.face_locations(gray, number_of_times_to_upsample=0, model="cnn")

        if len(face_locations) > 0:
            # Show the original frame with face rectangles drawn around the faces
            for top, right, bottom, left in face_locations:
                # cv2.rectangle(frame, (left, top), (right, bottom), (0, 0, 255), 2)
                face_image = gray[top:bottom, left:right]

                # Resize the face image to the desired size
                resized_face_image = cv2.resize(face_image, (128,128))

                # Predict the class of the resized face image using the model
                result = model.predict(resized_face_image)
                print(result[0])
                if(result[0] == 'on_camera'): on_camera = on_camera + 1
                elif(result[0] == 'off_camera'): off_camera = off_camera + 1
                total = total + 1
            
        # cv2.imshow('Video', frame)

        # If the user presses the 'q' key, exit the loop
        # if cv2.waitKey(1) & 0xFF == ord('q'):
        #     break
    gaze_percentage = on_camera/total*100
    # print(total,on_camera,off_camera)
    # print(f'focus perfectage = {on_camera/total*100}')
    # Release the video capture object and close all windows
    video_capture.release()
    cv2.destroyAllWindows()
    end_time = time.time()
    print(f'Time taken: {end_time-start_time}')
    return gaze_percentage


demo = gr.Interface(fn = video_processing,
                    inputs= gr.Video(),
                    outputs = gr.Text()
                    )

if __name__ == "__main__":
    demo.launch()