Urban-Elements-ReID---baseline

In this repository, you can find instructions on how to download, configure, and run the baseline for the Urban Elements ReID competition.

Download code and set up enviroment

To download the main code and set up the environment, please follow the instructions only up to downloading the pre-trained model (step 3) of Part Aware Transformer.

Modified codes

In order to use PAT for the Urban Elements ReID competition follow the next steps:

1) Download the UrbanElementsReID dataset

Download the UrbanElementsReID dataset from the section Data in the Kaggle competition page and place it in the directory of your choice.

Once the dataset is downloaded (If the file structure does not look right, please refer to setup.sh.), run the setup.sh script over the dataset directoy in order to place the folders in the correct way. You can find this script in /Codes/setup.sh

cd "your data directory"

bash setup.sh

If needed give permissions to access the folders running

chmod +x "folder name"

2) Add the required files

Add to the folder Part-Aware-Transformer/data/datasets/ the dataloaders and initialization files UrbanElementsReID.py, UrbanElementsReID_test.py and __init__.py.

Add to Part-Aware-Transformer/config/ folder and set up the correspondig paths and configuration of UrbanElementsReID_test.yml and UrbanElementsReID_train.yml files.

Add to Part-Aware-Transformer/utils/ the file re_rankig.py.

Add to Part-Aware-Transformer/ the evaluation files update.py and evaluate_csv.py

3) Set up configuration files

Modify the configuration files UrbanElementsReID_test.yml and UrbanElementsReID_train.yml and set up your path to the data directory (DATASET:ROOT_DIR), pretrained model weigths (MODEL:PRETRAIN_PATH and TEST:WEIGHT) and output directory.

4) Train the model

In order to train the model first make sure that all the configuration settings and paths are correct. Then run the following line:

python train.py --config_file "config/UrbanElementsReID_train.yml"

5) Inference (Result Track Submission Generation)

To generate the final ranking for the competition, use the update.py script. This will process the test images and create a track_submission.csv file containing the top 100 candidates for each query. This file is what you typically upload to Kaggle or use for final scoring.

python update.py --config_file "config/UrbanElementsReID_test.yml" --track "path to store the files/track"

6) Local Evaluation (mAP & CMC Metrics)

If you want to calculate the performance of your model locally (without uploading to Kaggle) in the UAM dataset external data, use the evaluation script. This script compares your generated track_submission.csv against the ground truth (query.csv and test.csv) to calculate Mean Average Precision (mAP) and Cumulative Match Characteristic (CMC) curves.

python evaluate_csv.py --path "path/to/dataset/csv_folder/" --track "path/to/track_submission.csv"

References

[1]. Paula Moral, Alvaro García-Martín, Jose M. Martínez. (2024). Long-term geo-positioned re-identification dataset of urban elements. IEEE International Conference on Image Processing (ICIP), pp. 124-130, https://10.1109/ICIP51287.2024.10647759.

[2]. Javier Galán, Miguel González, Paula Moral, Álvaro García-Martín, José M. Martínez. (2025). Transforming urban waste collection inventory: AI-Based container classification and Re-Identification. Waste Management, Volume 199, pp. 25-35, https://doi.org/10.1016/j.wasman.2025.02.051.

Acknowledgment

Special thanks to liyuke65535 for the creation and publication of Part Aware Transformer repository and congratulations for the excelent work.

This work has been supported by the Ministerio de Ciencia, Innovación y Universidades of the Spanish Government under project SEGA-CV (TED2021-131643A-I00)