TAMPAR

This repo was developed as part of our WACV '24 paper "TAMPAR: Visual Tampering Detection for Parcels Logistics in Postal Supply Chains" (citation), also check the project page for more details.

Figure: We detect tampering by comparing the full parcel texture from a database (a) with the viewpoint-invariant parcel side surfaces of a single image by exploiting parcel corner point predictions (b). Appearance change detection is performed for each pair of matching parcel side surfaces to identify tampering (c). © IEEE 2024.

Usage

Setup

We highly recommend to use the provided Devcontainer to make the usage as easy as possible:

Install Docker and VS Code
Install VS Code Devcontainer extension ms-vscode-remote.remote-containers

Clone the repository

git clone https://github.com/a-nau/tampar.git

Press F1 (or CTRL + SHIFT + P) and select Dev Containers: Rebuild and Reopen Container
Go to Run and Debug (CTRL + SHIFT + D) and press the run button, alternatively press F5

Afterwards

Download the pre-trained SimSaC weights from here and paste them into src/simsac/weight
Download the pre-trained keypoint detection weights from here

Keypoint Detection

To run a training on our 5 example images run:

python src/tools/train_maskrcnn.py --config-file ./src/maskrcnn/configs/test.yaml --gpus "0" --num-gpus 1 --num-machines 1

To add your own dataset, you need to register in the register_datasets.py.
To check your results qualitatively you can use detectron_qualitative_evaluation.ipynb

Predict Tampering

We first need to compute all relevant similarity scores

python src/tools/compute_similarity_scores.py

Afterwards, we can train the decision tree and predict tampering using

python src/tools/predict_tampering.py

Note: This will run only on the sample data from data/tampar_sample/.

TAMPAR dataset

You can download the dataset from Zenodo.

Figure: Visual samples from TAMPAR. Check our project website for more.

Citation

If you use the code of our paper for scientific research, please consider citing

@inproceedings{naumannTAMPAR2024,
    author    = {Naumann, Alexander and Hertlein, Felix and D\"orr, Laura and Furmans, Kai},
    title     = {TAMPAR: Visual Tampering Detection for Parcels Logistics in Postal Supply Chains},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
    month     = {January},
    year      = {2024},
    note      = {to appear in}
}

Affiliations

Credits

We use SimSaC inference code. Licensed under GPL-3.0 which applies to folder src/simsac
The Mask R-CNN is borrowed from CubeRefine R-CNN (See License) and this implementation (MIT License), which applies to folder src/maskrcnn

Unless otherwise stated, this repo is distributed under MIT License.