small changes for running statistics

script/smiling_detection_improved.py

@@ -1,15 +1,18 @@
 import numpy as np
 import cv2
+import os
 import time
 from face_recognition import face_locations
 import keras
 from keras.models import model_from_json
 from skimage.measure import block_reduce
 from skimage.transform import resize
 
-def liveCam(model):
- video = cv2.VideoCapture('/dev/video0')
- while True:
+def liveCam(model, video):
+ video = cv2.VideoCapture(video)
+ smile = 0.0
+ frames = video.get(cv2.CAP_PROP_FRAME_COUNT)
+ for i in range(int(video.get(cv2.CAP_PROP_FRAME_COUNT))):
  # Capture frame-by-frame
  #time.sleep(1)
  ret, frame = video.read()
@@ -25,9 +28,9 @@ def liveCam(model):
  head = head.astype(np.float)
  head = np.array([[head]])
  head = np.transpose(head, [0, 2, 3, 1])
- print("Working")
  probabilities = model.predict(head)[0]
- print("Results: {}".format(probabilities))
+ if probabilities[1] > smile:
+ smile = probabilities[1]
  # Display the resulting frame
  image = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
  cv2.imshow('Video', frame)
@@ -36,10 +39,23 @@ def liveCam(model):
  # When everything is done, release the capture
  video.release()
  cv2.destroyAllWindows()
+ return smile
 
 def main():
  model = model_from_json(open('models/model_smiling.json').read())
  model.load_weights('models/weights_smiling.h5')
+ for filename in os.listdir('data/l2torclips/small'):
+ if filename.endswith(".mp4"):
+ print(filename)
+ try:
+ smile_pred = liveCam(model,'data/l2torclips/small/' + filename)
+ print(smile_pred)
+ stat = "{}, {}\n".format(filename, smile_pred)
+ with open("smiling.txt", "a") as gz:
+ gz.write(stat)
+ except ValueError as err:
+ continue
+
  liveCam(model)
 
 if __name__ == '__main__':