This repo contains the source codes implemented to run the experiments for person re-identification used within the paper: "Aggregating Deep Pyramidal Representations for Person Re-Idenfitication", published in International Conference on Computer Vision and Pattern Recognition - Workshop on Target Re-identification and Multi-Target Multi-Camera Tracking, 2019.
The repository does not contain the datasets. You can download a copy of the Market-1501 dataset from here: Datasets. Just extract the zip within the "data" folder. To make the scripts running with other datasets (e.g., Duke, CUHK, etc.), you can just copy the original files with the same "data" folder.
The solution has been written using the PyTorch framework and tested with the version specified with the requirements.txt file. If you want, feel free to run
pip install -r requirements.txt
to get all the dependencies in place.
After that, you can just run
python main.py
to perform a train/test with single shot on the Market-1501 dataset (provided you have downloaded and copied it as described before).
If you want to test the solution with a different configuration, please have a look at the arguments within the main.py file. Those should be self-explanatory.
If you use the code contained in this package we appreciate if you'll cite our work.
BIBTEX: @inproceedings{Martinel2019a, author = {Martinel, Niki and Foresti, Gian Luca and Micheloni, Christian}, booktitle = {International Conference on Computer Vision and Pattern Recognition Workshops}, title = {{Aggregating Deep Pyramidal Representations for Person Re-Identification}}, year = {2019} }