app.py

import gradio as gr
import os
import cv2
import face_recognition
from fastai.vision.all import load_learner
import time
import base64
from deepface import DeepFace
import torchaudio, torch
import subprocess

# import pathlib
# temp = pathlib.PosixPath
# pathlib.PosixPath = pathlib.WindowsPath

backends = [
  'opencv', 
  'ssd', 
  'dlib', 
  'mtcnn', 
  'retinaface', 
  'mediapipe'
]

model = load_learner("gaze-recognizer-v3.pkl")

def video_processing(video_file, encoded_video):
    angry = 0
    disgust = 0
    fear = 0
    happy = 0
    sad = 0
    surprise = 0
    neutral = 0
    emotion_count = 0

    if encoded_video != "":
    
        decoded_file_data = base64.b64decode(encoded_video)

        with open("temp_video.mp4", "wb") as f:
            f.write(decoded_file_data)
        
        video_file = "temp_video.mp4"

    start_time = time.time()

    subprocess.call(["ffmpeg", "-i", 'temp_video.mp4', "audio.wav"])
    waveform, sample_rate = torchaudio.load("audio.wav")
    waveform, sample_rate

    waveform, sample_rate = torchaudio.load("audio.wav")
    resampler = torchaudio.transforms.Resample(sample_rate, 16000)
    waveform = resampler(waveform)[0]

    input_features = processor(waveform.squeeze(dim=0), return_tensors="pt").input_features 
    predicted_ids = model.generate(input_features)
    
    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
    print(transcription[0])
    
    video_capture = cv2.VideoCapture(video_file)
    on_camera = 0
    off_camera = 0
    total = 0

    while True:
        # Read a single frame from the video
        for i in range(24*3):
            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]
                color_image = frame[top:bottom, left:right]

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

                try:
                    emotion = DeepFace.analyze(color_image,actions=['emotion'],detector_backend = backends[2],enforce_detection = False)# 2,3, 4 works
                    emotion_count += 1
                except Exception as e:
                    pass

                print(emotion[0]['emotion'])
                angry += emotion[0]['emotion']['angry']
                disgust += emotion[0]['emotion']['disgust']
                fear += emotion[0]['emotion']['fear']
                happy += emotion[0]['emotion']['happy']
                sad += emotion[0]['emotion']['sad']
                surprise += emotion[0]['emotion']['surprise']
                neutral += emotion[0]['emotion']['neutral']

                # 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

    try:
        # your processing code here
        gaze_percentage = on_camera / total * 100
    except Exception as e:
        print(f"An error occurred while processing the video: {e}")
        gaze_percentage = f'no face detected Total = {total},on_camera = {on_camera},off_camera = {off_camera}'
    print(f'Total = {total},on_camera = {on_camera},off_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}')
    if os.path.exists("temp_video.mp4"): 
        os.remove("temp_video.mp4")
    print(gaze_percentage)

    angry = angry / emotion_count
    disgust = disgust / emotion_count
    fear = fear / emotion_count
    happy = happy / emotion_count
    sad = sad / emotion_count
    surprise = surprise / emotion_count
    neutral = neutral / emotion_count
    emotion = {
        'angry': angry, 
        'disgust': disgust, 
        'fear': fear, 
        'happy': happy, 
        'sad': sad, 
        'surprise': surprise, 
        'neutral': neutral
     }, 
    
    # angry = 'total anger percentage' + str(angry) 
    # disgust = 'total disgust percentage' + str(disgust) 
    # fear = 'total fear percentage' + str(fear) 
    # happy = 'total happy percentage' + str(happy) 
    # sad = 'total sad percentage' + str(sad) 
    # surprise = 'total surprise percentage' + str(surprise)
    # neutral = 'total neutral percentage' + str(neutral) 
    print(f'total anger percentage = {angry}')
    print(f'total disgust percentage = {disgust}')
    print(f'total fear percentage = {fear}')
    print(f'total happy percentage = {happy}')
    print(f'total sad percentage = {sad}')
    print(f'total surprise percentage = {surprise}')
    print(f'total neutral percentage = {neutral}')
    return str(gaze_percentage)


demo = gr.Interface(fn=video_processing,
                     inputs=["video", "text"],
                     outputs="text")

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