Elamur Research⌕
☰
home / papers / 2026 / Free-Gaza at NakbaArchiveClassifier Shared Task: Towards Distinguishing the Destructive Effect of Nakba: NakbaImage Classification Using Artificial Intelligence Techniques

Free-Gaza at NakbaArchiveClassifier Shared Task: Towards Distinguishing the Destructive Effect of Nakba: NakbaImage Classification Using Artificial Intelligence Techniques

Paper · 2026 · Nisreen I. R. Yassin and Enas A. Hakim Khalil

Summary

The paper describes the Free-Gaza team's system for the NakbaArchiveClassifier Shared Task 2026. The task is to label social-media photos from Gaza, taken since October 2023, as showing either destroyed or intact infrastructure. Manual sorting of such images does not scale, so automatic labelling would help archivists and researchers index this visual evidence.

The system had to run on a GPU with only 2GB of VRAM. Under that limit, deeper backbones such as ConvNeXt and DenseNet121 lost accuracy because they needed small batches. The authors therefore used two ImageNet-pretrained EfficientNetB0 models at 224×224 resolution:

  • Model A was trained on the task data with standard binary cross-entropy.
  • Model B was trained on a larger set (task data, the AIDER aerial disaster dataset, high-confidence pseudo-labels and the earlier validation split) with focal loss and a cosine-decay learning rate.

At inference, 8-way test-time augmentation feeds a weighted soft vote between the two models. The ensemble reached a macro F1 of 0.8416 on the official test set and ranked 9th. This shows that a modest, reproducible pipeline can work under tight hardware limits.

Key points

  • Task data: 1,400 training, 199 validation and 402 test PNG images, with about 35% in the destruction class.
  • Model A (validation macro F1 0.80) tended to predict "not destruction" and missed 23 of 70 destroyed images (recall 0.67).
  • Model B reached recall 0.97 for destruction and macro F1 0.85. However, it was validated on a different 1,090-image set drawn from AIDER, so its scores are not directly comparable with Model A's.
  • Ensemble: P = 0.40·P(A) + 0.60·P(B), averaged over 8 augmented views (flips, rotations, brightness/contrast changes).
  • Official test results: accuracy 0.8557, precision 0.8427, recall 0.8405 and F1 0.8416, ranking 9th. The system was implemented in TensorFlow.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksComputer Vision and Pattern Recognition · 2,869 worksSocial Sciences · 1,772 worksSociology and Political Science · 577 works

💡 Concepts 1Key ideas and methods explained on our concept pages.Artificial intelligence · 49 works

Cite

Cite as Nisreen I. R. Yassin and Enas A. Hakim Khalil. 2026. Free-Gaza at NakbaArchiveClassifier Shared Task: Towards Distinguishing the Destructive Effect of Nakba: NakbaImage Classification Using Artificial Intelligence Techniques. Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/59i3kmpytngj

@inproceedings{yassin2026free,
  title = {{Free-Gaza at NakbaArchiveClassifier Shared Task: Towards Distinguishing the Destructive Effect of Nakba: NakbaImage Classification Using Artificial Intelligence Techniques}},
  author = {Nisreen I. R. Yassin and Enas A. Hakim Khalil},
  booktitle = {Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026},
  year = {2026},
  doi = {10.63317/59i3kmpytngj},
  url = {https://doi.org/10.63317/59i3kmpytngj}
}

References

9 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Alexei Abrahams, Shadi Abudalfa, Mustafa Jarrar, and George Mikros. 2026. The nakbaarchiveclassifier shared task on nakba image classification. In Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources (Nakba-NLP 2026), co-located with the Language Resources and Eval… find
  2. Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, IEEE, pp. 248-255. find
  3. Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770-778.
  4. Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton. 2012. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25.
  5. Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. 2017. Focal loss for dense object detection. In Proceedings of the IEEE international conference on computer vision, pp. 2980-2988. *Figure 2: Confusion matrix for model B.* accuracy precision recall F1-score specificity Balance…
  6. The Ilya Loshchilov and Frank Hutter. 2016. Sgdr: Stochastic gradient descent with warm restarts. arXiv preprint arXiv:1608.03983. find
  7. Connor Shorten and Taghi M. Khoshgoftaar. 2019. A survey on image data augmentation for deep learning. Journal of big data, 6, pp. 1-48. find
  8. Mingxing Tan and Quoc Le. 2019. Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning, PMLR, pp. 6105-6114.
  9. Zhi-Hua Zhou. 2025. Ensemble methods: foundations and algorithms. Chapman and Hall/CRC. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yassin-2026-free-gaza-nakbaarchiveclassifier-shared-task/