Keras Sequence Video generators

This package proposes some classes to work with Keras (included in TensorFlow) that generates batches of frames from video files.

It is useful to work with Time Distributed Layer with Video input (frames). Then you may use GRU or LSTM. See articles:

• The basics of Video frame as input
• Usage of these generators here

Example can be seen in [integrated note book](./Example of usage.ipynb). To see the example in a proper viewer, you can use that URL.

Requirements are:

• Python3 (Python 2 will never been supported)
• OpenCV
• numpy
• Keras
• TensorFlow (or other backend, not tested)

If you want to compile the package, you need:

• sphinx to compile doc
• setuptools

Installation

You can install the package via pip:

pip install keras-video-generators

If you want to build from sources, clone the repository then:

python setup.py build

Usage

The package contains 3 generators that inherits Sequence interface. So they may be used with model.fit_generator():

• VideoFrameGenerator that will take the choosen number of frames from the entire video
• SlidingFrameGenerator that takes frames with decay for the entire video or with a sequence time
• OpticalFlowGenerator that gives optical flow sequence from frames with different methods

Each of these generators accepts parameters:

• glob_pattern that must contain {classname}, e.g. './videos/{classname}/*.png' - the "classname" in string is used to detect classes
• nb_frame that is the number of frame in the sequence
• batch_size
• transformation that can be None or or ImageDataGenerator to make data augmentation
• use_frame_cache to use with caution, if set to True, the class will keep frames in memory (without augmentation). You need a lot of memory
• and many more, see class documentation

Work in progress

• Complete documentation
• use Optical Flow on SlidingFrameGenerator