PoTATO - A Dataset for Analyzing Polarimetric Traces of Afloat Trash Objects

Description

This repository contains the scripts and code for the paper PoTATO: A Dataset for Analyzing Polarimetric Traces of Afloat Trash Objects presented during the ECCV24 TRICKY workshop. It allows for easy replication of the experiments and supports research reproducibility.

The PoTATO dataset contains more than 12k labeled plastic bottles with RAW polarimetric and color images.

Extracted Channels

Example of channels that can be extracted from the RAW images.

---

Example bottles in the dataset

Example of bottle images and the different channels that can be extracted.

Bottles used in the experiment.

---

Detection Example

The sequence of images illustrates the object detection that can be implemented independently each channel.

---

Steps for reproducing the experiments

1. Clone this repository:

git clone https://github.com/luisfelipewb/PoTATO/tree/eccv2024

2. Download the PoTATO dataset from dataset link:

mv ~/Downloads/potato.tgz ./datasets
cd ./datasets && tar -xzvf potato.tgz

3. Extract the images from the RAW data (This might take a while and will need 25GB space). The imagess will be generated in the same folder as the raw files:

cd ../
python3 utils/extract_from_raw.py -d datasets/potato/images_raw/

4. Place images in the correct folder:

find ./datasets/potato/images_raw/ -type f ! -name '*_raw.png' -exec mv {} ./datasets/potato/images/ \;

5. Install YOLO-v5: Same instructions from the YOLOv5 repository https://github.com/ultralytics/yolov5 The 'yolov5' folder should be on the same level as the 'experiments' folder

git clone https://github.com/ultralytics/yolov5
pip install -r ./yolov5/requirements.txt

6. Run YOLO training and detection:

cd ./experiments/
./train_yolo.sh

7. Install Detectron2 and remaining dependencies.

pip install r ./requirements.txt

8. Run Detectron2 training:

./train_det2.sh

9. Fix the image_id in yolo detection. The image_id used for the YOLO detections does no correspond to the annotation. The following script updates the image_id for extracting the COCO metrics

cd utils
python3 fix_yolo_detections.py

10. Generate metrics

python3 coco_metrics.py

The results will be avaiable in utils/output/metrics.csv

Please cite the paper when using the dataset or code

@inproceedings{batista2024potato,
      title={PoTATO: A Dataset for Analyzing Polarimetric Traces of Afloat Trash Objects},
      author={Batista, Luis F. W. and Khazem, Salim and Adibi, Mehran and Hutchinson, Seth and Pradalier, Cedric},
      year={2025},
      editor={Del Bue, Alessio and Canton, Cristian and Pont-Tuset, Jordi and Tommasi, Tatiana},
      booktitle={Computer Vision -- ECCV 2024 Workshops},
      publisher={Springer Nature Switzerland},
      address={Cham},
      pages={190--205},
      isbn={978-3-031-91569-7}
}