Signal, Image and Video Project

Report

Follow this read.me to understand how to execute our Signal, Image and Video project on your machine.

Environment setup

We suggest creating a virtual environment to install all the required Python packages in. Use the following commands to install Virtuslenv and create a new environment.

python3 -m pip install virtualenv
python3 -m venv /path/to/virtual/environment/env_name 

To start the environment

source env_name/bin/activate #on unix systems
source env_name/Scripts/activate #on Windows

To exit the environment

deactivate

To install the required packages use the reference file

python3 -m pip install -r requirementsCUDA.txt #for CUDA compatible systems
python3 -m pip install -r requirementsMAC.txt #for macOS silicon systems

If you have a GPU (higly reccomended) install torch with hardware acceleration. It is mandatory beacuse otherwise the code does not work.

Installation

To install the OpenReID library with our modifications and tricks, execute the following command in the main directory of the project. Please not that this operation is required every time a change to the project is done.

python3 setup.py install

Execution

To train the model with every trick just use

cd tricks
python3 triplet_loss.py -t 6 --combine-trainval

Use the following options to modify the settings

-t _ #to select the tricks to use (up to that number)
-d _ #to select the dataset to use
-b _ #to select the batch the size
--epochs _ #to change the number of epochs
--num-instances _ #to select the number of image per identity
--cross-domain #to make the final test on the other dataset (dukemtmc / market1501)
--data-dir _ #to change the dataset directory
--logs-dir _ #to change the logs directory
--evaluate #to just execute the evaluation
--resume _ #to resume training from the given checkpoint


-j _ #to select the number of workers
--height _ #to change the height of the input images
--width _ #to change the width of the input images
--re-ranking #to use re-ranking
--combine-trainval #to use validation images during training

Market1501 and DukeMTMC-reID can be downloaded here.