Plugin package: Object Detection and Tracking (ODT)

This package includes all methods to detect and track objects within on image sequence. This repo is based and is adapted on the following repositories:

https://github.com/eriklindernoren/PyTorch-YOLOv3
https://github.com/ZQPei/deep_sort_pytorch

Package Description

PDF format: vhh_od_pdf

HTML format (only usable if repository is available in local storage): vhh_od_html

Quick Setup

This package includes a setup.py script and a requirements.txt file which are needed to install this package for custom applications. The following instructions have to be done to used this library in your own application:

Requirements:

• Ubuntu 18.04 LTS
• CUDA 10.1 + cuDNN
• python version 3.6.x

We developed and tested this module with pytorch 1.8-1+cu111 and torchvision 0.9.1+cu111.

0 Environment Setup (optional)

Create a virtual environment:

• create a folder to a specified path (e.g. /xxx/vhh_od_env/)
• python3 -m venv /xxx/vhh_od_env/

Activate the environment:

• source /xxx/vhh_od_env/bin/activate

1A Install using Pip

The VHH Object Detection and Tracking package is available on PyPI and can be installed via pip.

• Update pip and setuptools (tested using pip==20.2.3 and setuptools==50.3.0)
• pip install vhh-od

Alternatively, you can also build the package from source.

1B Install by building from Source

Checkout vhh_stc repository to a specified folder:

• git clone https://github.com/dahe-cvl/vhh_od

Install the stc package and all dependencies:

• Update pip and setuptools (tested using pip==20.2.3 and setuptools==50.3.0)
• Install the wheel package: pip install wheel
• change to the root directory of the repository (includes setup.py)
• python setup.py bdist_wheel
• The aforementioned command should create a /dist directory containing a wheel. Install the package using python -m pip install dist/xxx.whl

NOTE: You can check the success of the installation by using the commend pip list. This command should give you a list with all installed python packages and it should include vhh-stc.

2 Install PyTorch

Install a Version of PyTorch depending on your setup. Consult the PyTorch website for detailed instructions.

3 Setup environment variables (optional)

• source /data/dhelm/python_virtenv/vhh_od_env/bin/activate
• export CUDA_VISIBLE_DEVICES=1
• export PYTHONPATH=$PYTHONPATH:/XXX/vhh_od/:/XXX/vhh_od/Develop/:/XXX/vhh_od/Demo/

4 Run demo script (optional)

* Make sure to have a video (e.g vid.m4v) in the videos folder (see "PATH_VIDEOS" in the config file) and a corresponding shot type detection result (e.g. vid.csv) in the stc results folder (see "STC_RESULTS_PATH" in the config file). 
* make sure that the vhh_od directory is in your Python-Path
* Settings can be adjusted via config/config_vhh_od.yaml
* ```cd Demo```
* ```python run_od_on_single_video.py```

5 Visualization (optional)

* Make sure to have a video (e.g vid.m4v) stored in the videos folder (see "PATH_VIDEOS" in the config file) and a corresponding object detection result in the results folder (see "PATH_FINAL_RESULTS" in the config file).
* Make sure that the vhh_od directory is in your Python-Path
* Settings can be adjusted via config/vis_config.yaml
* ```python Demo/run_visualization_on_single_video.py vid.m4v```
* This will create a video with bounding boxes in the raw results folder specified in config_vhh_od.yaml

Release Generation

* Create and checkout release branch: (e.g. v1.1.0): ```git checkout -b v1.1.0```
* Update version number in setup.py
* Update Sphinx documentation and release version
* Make sure that ```pip``` and ```setuptools``` are up to date
* Install ```wheel``` and ```twine```
* Build Source Archive and Built Distribution using ```python setup.py sdist bdist_wheel```
* Upload package to PyPI using ```twine upload dist/*```