python application to run FER on a dataset of images & extract/process a dataset of videos.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 

167 lines
4.4 KiB

from deepface import DeepFace
import os
import glob
import json
import numpy as np
baseFolderPath = "main_dataset"
lostFolderPath = "lost-images"
output_file = "sv_embeddings.json"
embeddings = []
backends = [
'opencv',
'ssd',
'dlib',
'mtcnn',
'fastmtcnn',
'retinaface',
'mediapipe',
'yolov8',
'yunet',
'centerface',
]
def main():
video_folders = setup()
for folder_info in video_folders:
foldername, video_path, image_paths = folder_info
print(f"Processing {foldername}\n")
run(image_paths, foldername, video_path)
saveToJson()
def processImage(image, idx, foldername, video_path):
try:
print(f"Processing {image}", end="\r")
# Initialize variables to accumulate sums
sums = {
'angry': 0,
'disgust': 0,
'fear': 0,
'happy': 0,
'sad': 0,
'surprise': 0,
'neutral': 0
}
anaylse_obj = DeepFace.analyze(
img_path=image,
actions=['emotion'],
enforce_detection=True,
detector_backend=backends[5],
silent=True
)
emotions = anaylse_obj[0]['emotion']
selected_emotions = [
emotions['happy'],
emotions['sad'],
emotions['angry'],
emotions['neutral']
]
num_of_faces = len(anaylse_obj)
# Get average emotion
# for face in anaylse_obj:
# emotions = face['emotion']
# for emotion in sums:
# sums[emotion] += emotions.get(emotion, 0)
# averages = {emotion: sums[emotion] / num_of_faces for emotion in sums}
raw_emotions = [value for value in emotions.values()]
#normalized_emotions = [value / 100 for value in raw_emotions]
normalized_emotions = [value / 100 for value in selected_emotions]
#normalised_emotions = l2_normalize(np.array(raw_emotions))
# normalised_emotions = normalised_emotions.tolist()
entry = {
"folder": foldername,
"image": image, # Extract the filename from the path
"error": 0,
"vector": normalized_emotions
}
# Add the entry to the JSON data list
embeddings.append(entry)
except:
# if model does not detect any emotions
# set lost -> 1 & generate a random emotion vector
random_vector = np.random.rand(4)
random_vector_list = random_vector.tolist()
entry = {
"folder": foldername,
"image": image, # Extract the filename from the path
"error": 1,
"vector": random_vector_list
}
# Add the entry to the JSON data list
embeddings.append(entry)
pass
print(entry["vector"])
def setup():
video_folders = []
subfolders = [f.path for f in os.scandir(baseFolderPath) if f.is_dir()]
for subfolder in subfolders:
frames_folder_path = os.path.join(subfolder, "segmented_images")
# Find the video file in the subfolder
# video_files = [f for f in os.listdir(subfolder) if f.endswith(('.mp4', '.avi', '.mov'))]
# if video_files:
# video_path = os.path.join(subfolder, video_files[0])
# List of all image files in the frames folder
image_files = glob.glob(os.path.join(frames_folder_path, '*'))
# Filter by specific image extensions (e.g., .jpg, .png)
image_paths = [img for img in image_files if img.endswith(('.jpg', '.jpeg', '.png', '.bmp', '.tiff', '.gif'))]
# Sort the image paths by filename (assuming the filenames are numerically ordered)
image_paths.sort(key=lambda x: int(os.path.splitext(os.path.basename(x))[0]))
video_path = "null"
# Add the information to the list
video_folders.append((os.path.basename(subfolder), video_path, image_paths))
return video_folders
def run(image_paths, foldername, video_path):
for idx, image in enumerate(image_paths):
processImage(image, idx, foldername, video_path)
def l2_normalize(vector):
norm = np.linalg.norm(vector)
if norm == 0:
return vector
return vector / norm
def saveToJson():
with open(output_file, 'w') as f:
json.dump(embeddings, f, indent=4)
print("Successfully created JSON file.")
if __name__ == "__main__":
main()