📘 Evaluating Temporal Context for Robustness to Perturbations in Video Object Detection Models - Results

Context & Methodology

This dataset was generated as part of the research project “Evaluating Temporal Context for Robustness to Perturbations in Video Object Detection Models”, focusing on video-based drone detection using temporal attention networks.

DMP DOI: 10.5281/zenodo.17771932

The objective of the dataset is to:
- Evaluate the robustness of different object detection architectures.
- Assess performance under varying noise levels.
- Investigate whether temporal aggregation improves reliability in degraded visual conditions.

Underlying datasets:
- VisDrone (CC BY-NC-SA 3.0)
- XS-VID (MIT License)

Code to create this dataset can be found under: https://github.com/mozi30/TemporalAttentionPlayground.git

Purpose & Reuse Scenarios

Dataset can be reused for robustness benchmarking, temporal inference analysis, or further drone vision research.

Technical DetailsFolder structure, naming conventions, and CSV fields are standardized. CSV format is used for interoperability.

Software Requirements

Python ≥3.8, PyTorch ≥1.10, CSV viewer for synthetic data otherwise read the README.md of code repository.

Reuse Notes

Noise robustness data simulated to illustrate expected performance patterns; see scripts.

Licensing

- VisDrone-based results: CC BY-NC-SA 3.0
- XS-VID-based results: MIT
- Code: MIT

---

See the README.md for additional information.

For full experiment setup and FAIR compliance, refer to the Data Management Plan (10.5281/zenodo.17771932) or repository: https://github.com/mozi30/TemporalAttentionPlayground#