Update for compatibility with newer versions of tensorflow and numpy

code/inference_preprocess.py

@@ -1,59 +1,71 @@
 import numpy as np
 import cv2
 from skimage.util import img_as_float
-import tensorflow as tf
 import matplotlib.pyplot as plt
-import time
-import scipy.io
 from scipy.sparse import spdiags
 
+
 def preprocess_raw_video(videoFilePath, dim=36):
 
     #########################################################################
     # set up
     t = []
     i = 0
-    vidObj = cv2.VideoCapture(videoFilePath);
-    totalFrames = int(vidObj.get(cv2.CAP_PROP_FRAME_COUNT)) # get total frame size
-    Xsub = np.zeros((totalFrames, dim, dim, 3), dtype = np.float32)
+    vidObj = cv2.VideoCapture(videoFilePath)
+    totalFrames = int(vidObj.get(cv2.CAP_PROP_FRAME_COUNT))  # get total frame size
+    Xsub = np.zeros((totalFrames, dim, dim, 3), dtype=np.float32)
     height = vidObj.get(cv2.CAP_PROP_FRAME_HEIGHT)
     width = vidObj.get(cv2.CAP_PROP_FRAME_WIDTH)
     success, img = vidObj.read()
-    dims = img.shape
     print("Orignal Height", height)
     print("Original width", width)
     #########################################################################
     # Crop each frame size into dim x dim
     while success:
-        t.append(vidObj.get(cv2.CAP_PROP_POS_MSEC))# current timestamp in milisecond
-        vidLxL = cv2.resize(img_as_float(img[:, int(width/2)-int(height/2 + 1):int(height/2)+int(width/2), :]), (dim, dim), interpolation = cv2.INTER_AREA)
-        vidLxL = cv2.rotate(vidLxL, cv2.ROTATE_90_CLOCKWISE) # rotate 90 degree
-        vidLxL = cv2.cvtColor(vidLxL.astype('float32'), cv2.COLOR_BGR2RGB)
+        t.append(vidObj.get(cv2.CAP_PROP_POS_MSEC))  # current timestamp in milisecond
+        vidLxL = cv2.resize(
+            img_as_float(
+                img[
+                    :,
+                    int(width / 2)
+                    - int(height / 2 + 1) : int(height / 2)
+                    + int(width / 2),
+                    :,
+                ]
+            ),
+            (dim, dim),
+            interpolation=cv2.INTER_AREA,
+        )
+        # vidLxL = cv2.rotate(vidLxL, cv2.ROTATE_90_CLOCKWISE) # rotate 90 degree
+        vidLxL = cv2.cvtColor(vidLxL.astype("float32"), cv2.COLOR_BGR2RGB)
         vidLxL[vidLxL > 1] = 1
-        vidLxL[vidLxL < (1/255)] = 1/255
+        vidLxL[vidLxL < (1 / 255)] = 1 / 255
         Xsub[i, :, :, :] = vidLxL
-        success, img = vidObj.read() # read the next one
+        success, img = vidObj.read()  # read the next one
         i = i + 1
     plt.imshow(Xsub[0])
-    plt.title('Sample Preprocessed Frame')
+    plt.title("Sample Preprocessed Frame")
     plt.show()
     #########################################################################
     # Normalized Frames in the motion branch
     normalized_len = len(t) - 1
-    dXsub = np.zeros((normalized_len, dim, dim, 3), dtype = np.float32)
+    dXsub = np.zeros((normalized_len, dim, dim, 3), dtype=np.float32)
     for j in range(normalized_len - 1):
-        dXsub[j, :, :, :] = (Xsub[j+1, :, :, :] - Xsub[j, :, :, :]) / (Xsub[j+1, :, :, :] + Xsub[j, :, :, :])
+        dXsub[j, :, :, :] = (Xsub[j + 1, :, :, :] - Xsub[j, :, :, :]) / (
+            Xsub[j + 1, :, :, :] + Xsub[j, :, :, :]
+        )
     dXsub = dXsub / np.std(dXsub)
     #########################################################################
     # Normalize raw frames in the apperance branch
     Xsub = Xsub - np.mean(Xsub)
-    Xsub = Xsub  / np.std(Xsub)
-    Xsub = Xsub[:totalFrames-1, :, :, :]
+    Xsub = Xsub / np.std(Xsub)
+    Xsub = Xsub[: totalFrames - 1, :, :, :]
     #########################################################################
     # Plot an example of data after preprocess
-    dXsub = np.concatenate((dXsub, Xsub), axis = 3);
+    dXsub = np.concatenate((dXsub, Xsub), axis=3)
     return dXsub
 
+
 def detrend(signal, Lambda):
     """detrend(signal, Lambda) -> filtered_signal
     This function applies a detrending filter.
@@ -80,5 +92,4 @@ def detrend(signal, Lambda):
     diags_data = np.array([ones, minus_twos, ones])
     diags_index = np.array([0, 1, 2])
     D = spdiags(diags_data, diags_index, (signal_length - 2), signal_length).toarray()
-    filtered_signal = np.dot((H - np.linalg.inv(H + (Lambda ** 2) * np.dot(D.T, D))), signal)
-    return filtered_signal
+    return np.dot((H - np.linalg.inv(H + (Lambda**2) * np.dot(D.T, D))), signal)