vdetlib - Python library for object detection in videos

Introduction

The vdetlib python library serves to detection objects in videos. It was originally developed for the ImageNet VID challenge introduced in ILSVRC2015. It contains components such as region proposal, still-image object detection, generic object tracking, spatial max-pooling and temporal convolution.

The T-CNN framework contains many tools that utilizes vdetlib. Please checkout that repository if you are interested.

Citing vdetlib

If you find vdetlib useful in your research and related project, please consider citing the following work accepted in CVPR 2016.

@inproceedings{kang2016object,
  Title = {Object Detection from Video Tubelets with Convolutional Neural Networks},
  Author = {Kang, Kai and Ouyang, Wanli and Li, Hongsheng and Wang, Xiaogang},
  Booktitle = {CVPR},
  Year = {2016}
}

License

This project is released under the MIT License.

Installations

Prerequisites

1. caffe with Python layer and pycaffe
2. FCN tracker
3. Matlab with python engine

Instructions

1. Clone the repository

       $ git clone https://github.com/myfavouritekk/vdetlib.git
   

2. Compilation

       $ cd vdetlib
       $ make
   

Protocols

There are some basic protocol types for using this library. All of them are defined as python dictionaries and are saved as JSON files. The definitions are written in the protocol.py.

To-do list

• [ ] detailed documentation
• [ ] demo script