keras-video-generators 1.0.14

Keras sequence generators for video data

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

An provided example of usage can be displayed in nbviewer here.

Requirements are:

• Python >= 3.6 (Python 2 will never been supported)
• OpenCV
• numpy
• Keras >= 2
• TensorFlow >= 1.15 (or other backend, not tested, TensorFlow is needed by Keras)

TensorFlow 2 works as well. This requirements is not integrated in the setup.py to let you choose the version, or to let you try with other backend. We mean that you will need to install a backend yourself (e.g. pip install tensorflow)

If you want to compile the package, you need:

• sphinx to compile doc (work in progress)
• 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 (experimental)

Each of these generators accepts parameters:

• glob_pattern that must contain {classname}, e.g. './videos/{classname}/*.avi' - the "classname" in string is used to detect classes
• nb_frames that is the number of frame in the sequence
• batch_size that is the number of sequence in one batch
• 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

Changelog

v1.0.14

• Changes to get first and last frames in sequence

v1.0.13

• try to fix SageMaker problem by avoiding usage of internal keras from tensorflow

v1.0.12

• fix transformation error with SlidingFrameGenerator

v1.0.11

• set generator to be Iterable
• frame cache was disabled by error, it's back now
• fixup import Sequence from tensorflow.keras
• fix frame count problems for video with bad headers

v1.0.10

• fix Windows problems with path using backslashes
• add auto discovery for classes if "None" is sent
• add travis tests

v1.0.9

• fix frame counter in SlidingFrameGenerator

v1.0.8

• fix tiny video frame count
• refactorisation
• pep8
• fix problems for video without headers

v1.0.7

• fix name check in classes to get them from filename
• add split_test and split_val

v1.0.5

• fix package generation

v1.0.4

• fix assertion