/video-matching

A submission on TRECVID 2016 video and caption matching task

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

A video and caption matching system (deprecated)

Introduction

Prerequisites

  • Python 2.7.x or 3.5.x

  • Keras installation with Theano or Tensorflow backend, but with tensorflow matrix order. If you are not sure how to setup, use the following configuration in your ~/.keras/keras.json:

{
    "image_dim_ordering": "tf",
    "epsilon": 1e-07,
    "floatx": "float32",
    "backend": "theano"
}
  • A configuration for Theano (use in ~/.theanorc, with GPU+CUDNN):
[global]
floatX = float32 # from cuda-8.0 can we use float16?
device = gpu
optimizer = fast_compile
mode = FAST_RUN

[dnn]
enabled = True

[lib]
cnmem = 1 # maybe it is good

Structure

Contact and future maintain

I actually do not have any plans to maintain or support this Python implementation in future (I'm moving all my codebase to Java now, sadly). For any technical discussions, please file an issue.