Use Tensorflow 2.5 now, go to 1.1 version

1.1 releases supports now tensorflow >=2.5

see #46

setup.py

@@ -1,27 +1,22 @@
 #!/usr/bin/env python
 
-from setuptools import setup, find_packages
+from setuptools import find_packages, setup
 
 with open("README.md", "r") as fh:
     long_description = fh.read()
 
 setup(
-    name='keras-video-generators',
-    version='1.0.13',
-    description='Keras sequence generators for video data',
+    name="keras-video-generators",
+    version="1.1.0",
+    description="Keras sequence generators for video data",
     long_description=long_description,
     long_description_content_type="text/markdown",
-    license='MIT',
-    licence_file='LICENSE',
-    author='Patrice Ferlet',
-    author_email='[email protected]',
-    url='https://github.com/metal3d/keras-video-generators',
-    package_dir={'': 'src'},
-    packages=find_packages('src'),
-    install_requires=[
-        'keras>=2',
-        'numpy',
-        'opencv-python',
-        'matplotlib'
-    ]
+    license="MIT",
+    licence_file="LICENSE",
+    author="Patrice Ferlet",
+    author_email="[email protected]",
+    url="https://github.com/metal3d/keras-video-generators",
+    package_dir={"": "src"},
+    packages=find_packages("src"),
+    install_requires=["tensorflow>=2.5", "numpy", "opencv-python", "matplotlib"],
 )

src/keras_video/flow.py

@@ -16,10 +16,11 @@
 
 """
 
-import numpy as np
 import cv2 as cv
+import numpy as np
+import tensorflow.keras.preprocessing.image as kimage
+
 from .generator import VideoFrameGenerator
-import keras.preprocessing.image as kimage
 
 METHOD_OPTICAL_FLOW = 1
 METHOD_FLOW_MASK = 2
@@ -57,15 +58,16 @@ class OpticalFlowGenerator(VideoFrameGenerator):
     """
 
     def __init__(
-            self,
-            *args,
-            nb_frames=5,
-            method=METHOD_OPTICAL_FLOW,
-            flowlevel=3,
-            iterations=3,
-            winsize=15,
-            **kwargs):
-        super().__init__(nb_frames=nb_frames+1, *args, **kwargs)
+        self,
+        *args,
+        nb_frames=5,
+        method=METHOD_OPTICAL_FLOW,
+        flowlevel=3,
+        iterations=3,
+        winsize=15,
+        **kwargs
+    ):
+        super().__init__(nb_frames=nb_frames + 1, *args, **kwargs)
         self.flowlevel = flowlevel
         self.iterations = iterations
         self.winsize = winsize
@@ -107,13 +109,20 @@ def make_optical_flow(self, images):
                 images[i] = cv.cvtColor(image, cv.COLOR_RGB2GRAY)
 
         flow = cv.calcOpticalFlowFarneback(
-            images[0], images[1],  # image prev and next
-            None, 0.5, self.flowlevel,
-            self.winsize, self.iterations,
-            5, 1.1, 0)
+            images[0],
+            images[1],  # image prev and next
+            None,
+            0.5,
+            self.flowlevel,
+            self.winsize,
+            self.iterations,
+            5,
+            1.1,
+            0,
+        )
 
         mag, ang = cv.cartToPolar(flow[..., 0], flow[..., 1])
-        hsv[..., 0] = ang*180/np.pi/2
+        hsv[..., 0] = ang * 180 / np.pi / 2
         hsv[..., 2] = cv.normalize(mag, None, 0, 255, cv.NORM_MINMAX)
         rgb = cv.cvtColor(hsv, cv.COLOR_HSV2BGR)
 
@@ -123,53 +132,54 @@ def make_optical_flow(self, images):
         return rgb
 
     def diff_mask(self, images):
-        """ Get absolute diff mask, then merge frames and apply the mask
-        """
+        """Get absolute diff mask, then merge frames and apply the mask"""
         mask = self.absdiff(images)
         mask = cv.GaussianBlur(mask, (15, 15), 0)
 
-        image = cv.addWeighted(images[0], .5, images[1], .5, 0)
+        image = cv.addWeighted(images[0], 0.5, images[1], 0.5, 0)
 
         return cv.multiply(image, mask)
 
     def flow_mask(self, images):
         """
         Get optical flow on images, then merge images and apply the mask
         """
-        mask = self.make_optical_flow(images) / 255.
+        mask = self.make_optical_flow(images) / 255.0
         mask = cv.GaussianBlur(mask, (15, 15), 0)
 
-        image = cv.addWeighted(images[0], .5, images[1], .5, 0)
+        image = cv.addWeighted(images[0], 0.5, images[1], 0.5, 0)
 
         return cv.multiply(image, mask)
 
     def get_validation_generator(self):
         """ Return the validation generator if you've provided split factor """
         return self.__class__(
             method=self.method,
-            nb_frames=self.nbframe-1,
+            nb_frames=self.nbframe - 1,
             nb_channel=self.nb_channel,
             target_shape=self.target_shape,
             classes=self.classes,
             batch_size=self.batch_size,
             shuffle=self.shuffle,
             rescale=self.rescale,
             glob_pattern=self.glob_pattern,
-            _validation_data=self.validation)
+            _validation_data=self.validation,
+        )
 
     def get_test_generator(self):
         """ Return the validation generator if you've provided split factor """
         return self.__class__(
             method=self.method,
-            nb_frames=self.nbframe-1,
+            nb_frames=self.nbframe - 1,
             nb_channel=self.nb_channel,
             target_shape=self.target_shape,
             classes=self.classes,
             batch_size=self.batch_size,
             shuffle=self.shuffle,
             rescale=self.rescale,
             glob_pattern=self.glob_pattern,
-            _test_data=self.test)
+            _test_data=self.test,
+        )
 
     def __getitem__(self, idx):
         batch = super().__getitem__(idx)
@@ -179,9 +189,9 @@ def __getitem__(self, idx):
             imgs = item
             batch_len = len(imgs)
             frames = []
-            for i in range(batch_len-1):
+            for i in range(batch_len - 1):
                 im1 = imgs[i]
-                im2 = imgs[i+1]
+                im2 = imgs[i + 1]
                 if self.method == METHOD_OPTICAL_FLOW:
                     image = self.make_optical_flow((im1, im2))
                 elif self.method == METHOD_ABS_DIFF: