---
title: "HCMUS_TheFangs at NakbaArchiveClassifier Shared Task:"
description: "This paper describes the top-ranked submission to the Nakba-NLP 2026 shared task called NakbaArchiveClassifier. 21 references."
authors: ["Duy Minh Dao Sy", "Trung Kiet Huynh", "Nguyen Chi Tran", "Phu Quy Nguyen Lam", "Phu Hoa Pham"]
year: 2026
venue: "Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026"
source: "https://aclanthology.org/2026.nakbanlp-1.15/"
doi: "10.63317/26tv58oaqfc4"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/sy-2026-hcmus-thefangs-nakbaarchiveclassifier-shared-task/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/sy-2026-hcmus-thefangs-nakbaarchiveclassifier-shared-task/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/32/) / HCMUS\_TheFangs at NakbaArchiveClassifier Shared Task: Foundation Models and Advanced Training Strategies for Conflict Damage Classification

# HCMUS\_TheFangs at NakbaArchiveClassifier Shared Task: Foundation Models and Advanced Training Strategies for Conflict Damage Classification

**Authors:** [Duy Minh Dao Sy](https://md.research.elamur.ai/free/en/authors/duy-minh-dao-sy/), [Trung Kiet Huynh](https://md.research.elamur.ai/free/en/authors/trung-kiet-huynh/), [Nguyen Chi Tran](https://md.research.elamur.ai/free/en/authors/nguyen-chi-tran/), [Phu Quy Nguyen Lam](https://md.research.elamur.ai/free/en/authors/phu-quy-nguyen-lam/) and [Phu Hoa Pham](https://md.research.elamur.ai/free/en/authors/phu-hoa-pham/) · **Year:** 2026 · **Venue:** [Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026](https://md.research.elamur.ai/free/en/venues/nakbanlp/) ([International Workshop on Nakba Narratives as Language Resources @ LREC 2026](https://md.research.elamur.ai/free/en/venues/nakbanlp/2026/)) · **Publisher:** [European Language Resources Association (ELRA)](https://md.research.elamur.ai/free/en/publishers/european-language-resources-association-elra/) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.nakbanlp-1.15/) · [PDF](https://aclanthology.org/2026.nakbanlp-1.15.pdf) · [DOI 10.63317/26tv58oaqfc4](https://doi.org/10.63317/26tv58oaqfc4) · [OpenAlex](https://openalex.org/W7166551073)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
This paper describes the top-ranked submission to the Nakba-NLP 2026 shared task called NakbaArchiveClassifier. The task supports the SaltPillar archive of Gaza conflict imagery. Systems label Instagram photos as either destruction or not_destruction, and submissions are scored by macro F1. The data pose three problems: the training set is small and imbalanced, the visual content is unlike ImageNet, and many photographers appear in both training and validation. That overlap lets a model succeed by learning an author's style instead of recognising damage.

Under one fixed training protocol, the authors compared five backbone families, among them EfficientNet-B4, DINOv2, EVA-02, UniRepLKNet and MaxViT. The best setup was MaxViT-Base with focal loss, MixUp and a six-step augmentation pipeline. Several of its augmentations were chosen specifically to remove photographer-specific cues.

This system reached a test macro F1 of 0.899 and placed first. The authors also report two approaches that did not work, a KAN head and BLIP-2 caption regularisation. The study shows that a good architectural fit and bias-aware augmentation matter more than model size when labelled data is scarce.

## Key points
- **Dataset:** 1,400 training, 199 validation and 402 test images, with a consistent 1.83:1 not_destruction/destruction ratio. 118 creators are shared between train and validation.
- **Leaderboard:** MaxViT-Base with focal loss (α=0.5, γ=1.5) and MixUp (p=0.5) scored 0.899 test macro F1. This was 0.004 above the second-placed team, and the top five teams were within 0.033 of each other.
- **Ablations (validation F1):**
  - Replacing focal loss with LDAM+DRW cost 0.134.
  - Replacing MaxViT with EfficientNet-B4 cost 0.052.
  - Removing MixUp cost 0.015.
- **Overfitting in larger models:** DINOv2-based runs showed the widest gaps between validation and test scores. The KAN head fell from 0.865 on validation to 0.552 on test. The authors attribute this collapse to the backbone and the spline parameters needing very different learning rates.
- **Caption regularisation:** Regularising with BLIP-2 captions through an InfoNCE loss scored only 0.804 on test, because the generic captions lacked conflict-specific detail. Code is released on GitHub (hcmus-thefangs).

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

## Institutions

- [Vietnam National University Ho Chi Minh City](https://md.research.elamur.ai/free/en/universities/vietnam-national-university-ho-chi-minh-city/) — University, Ho Chi Minh City, Vietnam · 21 works · 14 citations

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 works · [⚙️ Engineering](https://md.research.elamur.ai/free/en/shelves/engineering/) · 133 works · [⚙️ Media Technology](https://md.research.elamur.ai/free/en/shelves/engineering/media-technology/) · 34 works

## Cite

Cite as: Duy Minh Dao Sy, Trung Kiet Huynh, Nguyen Chi Tran, Phu Quy Nguyen Lam and Phu Hoa Pham. 2026. HCMUS\_TheFangs at NakbaArchiveClassifier Shared Task: Foundation Models and Advanced Training Strategies for Conflict Damage Classification. 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.

```bibtex
@inproceedings{sy2026hcmus,
  title = {{HCMUS\_TheFangs at NakbaArchiveClassifier Shared Task: Foundation Models and Advanced Training Strategies for Conflict Damage Classification}},
  author = {Duy Minh Dao Sy, Trung Kiet Huynh, Nguyen Chi Tran, Phu Quy Nguyen Lam and Phu Hoa Pham},
  booktitle = {Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026},
  year = {2026},
  doi = {10.63317/26tv58oaqfc4},
  url = {https://doi.org/10.63317/26tv58oaqfc4}
}
```

- [← previous HCMUS_TheFangs at AbjadGenEval Shared Task: Weighted Layer Pooling with Attention Fusion for Arabic AI-Generated Text Detection](https://md.research.elamur.ai/free/en/papers/sy-2026-hcmus-thefangs-abjadgeneval-shared-task/)
- [next → HCMUS_TheFangs at NakbaVirality Shared Task: The Audience is the Message: Escaping the Deep Learning Trap in Conflict-Domain Virality Prediction](https://md.research.elamur.ai/free/en/papers/sy-2026-hcmus-thefangs-nakbavirality-shared-task/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/32/)

## References

21 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. object detection. In Proceedings of the IEEE international conference on computer vision, pages 2980–2988. [find](https://scholar.google.com/scholar?q=object+detection%2E+In+Proceedings+of+the+IEEE+international+conference+on+computer+vision%2C+pages+2980%E2%80%932988%2E)
2. 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](https://scholar.google.com/scholar?q=Alexei+Abrahams%2C+Shadi+Abudalfa%2C+Mustafa+Jarrar%2C+and+George+Mikros%2E+2026%2E+The+nakbaarchiveclassifier+shared+task+on+nakba+image+classification%2E+In+Proceedings+of+the+2nd+International+Workshop+on+Nakb)
3. Alexander Buslaev, Vladimir I. Iglovikov, Eugene Khvedchenya, Alex Parinov, Mikhail Druzhinin, and Alexey A. Kalinin. 2020. Albumentations: Fast and flexible image augmentations. Information, 11(2):125. [find](https://scholar.google.com/scholar?q=Alexander+Buslaev%2C+Vladimir+I%2E+Iglovikov%2C+Eugene+Khvedchenya%2C+Alex+Parinov%2C+Mikhail+Druzhinin%2C+and+Alexey+A%2E+Kalinin%2E+2020%2E+Albumentations%3A+Fast+and+flexible+image+augmentations%2E+Information%2C+11%282%29%3A12)
4. Kaidi Cao, Colin Wei, Adrien Gaidon, Nikos Arechiga, and Tengyu Ma. 2019. Learning imbalanced datasets with label-distribution-aware margin loss. In Advances in Neural Information Processing Systems. [find](https://scholar.google.com/scholar?q=Kaidi+Cao%2C+Colin+Wei%2C+Adrien+Gaidon%2C+Nikos+Arechiga%2C+and+Tengyu+Ma%2E+2019%2E+Learning+imbalanced+datasets+with+label%2Ddistribution%2Daware+margin+loss%2E+In+Advances+in+Neural+Information+Processing+Systems%2E)
5. Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. ImageNet: A large-scale hierarchical image database. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 248–255. [find](https://scholar.google.com/scholar?q=Jia+Deng%2C+Wei+Dong%2C+Richard+Socher%2C+Li%2DJia+Li%2C+Kai+Li%2C+and+Li+Fei%2DFei%2E+2009%2E+ImageNet%3A+A+large%2Dscale+hierarchical+image+database%2E+In+Proceedings+of+the+IEEE+Conference+on+Computer+Vision+and+Pattern+R)
6. Xiaohan Ding, Yiyuan Zhang, Yixiao Ge, Sijie Zhao, Lin Song, Xiangyu Yue, and Ying Shan. 2024. Unireplknet: A universal perception large-kernel convnet for audio, video, point cloud, time-series and image recognition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti… [find](https://scholar.google.com/scholar?q=Xiaohan+Ding%2C+Yiyuan+Zhang%2C+Yixiao+Ge%2C+Sijie+Zhao%2C+Lin+Song%2C+Xiangyu+Yue%2C+and+Ying+Shan%2E+2024%2E+Unireplknet%3A+A+universal+perception+large%2Dkernel+convnet+for+audio%2C+video%2C+point+cloud%2C+time%2Dseries+and+i)
7. Yuxin Fang, Jiaqi Sun, Jianian Wang, Xiaosong Wang, Tiejun Huang, Xinlong Wang, and Yue Cao. 2023. Eva-02: A visual representation-level-mimicri transformer for image recognition. arXiv preprint arXiv:2303.11331. [find](https://scholar.google.com/scholar?q=Yuxin+Fang%2C+Jiaqi+Sun%2C+Jianian+Wang%2C+Xiaosong+Wang%2C+Tiejun+Huang%2C+Xinlong+Wang%2C+and+Yue+Cao%2E+2023%2E+Eva%2D02%3A+A+visual+representation%2Dlevel%2Dmimicri+transformer+for+image+recognition%2E+arXiv+preprint+arXiv)
8. Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, François Laviolette, Mario Marchand, and Victor Lempitsky. 2016. Domain-adversarial training of neural networks. volume 17, pages 2096–2030. [find](https://scholar.google.com/scholar?q=Yaroslav+Ganin%2C+Evgeniya+Ustinova%2C+Hana+Ajakan%2C+Pascal+Germain%2C+Hugo+Larochelle%2C+Fran%C3%A7ois+Laviolette%2C+Mario+Marchand%2C+and+Victor+Lempitsky%2E+2016%2E+Domain%2Dadversarial+training+of+neural+networks%2E+volume)
9. Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. 2023. [BLIP-2: Bootstrapping language-image pre-training with frozen image encoders and large language models](https://md.research.elamur.ai/free/en/papers/li-2023-blip-2-bootstrapping-language-image/). In International Conference on Machine Learning, pages 19730– 19742. PMLR.
10. Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. 2017. Focal loss for dense Ziming Liu, Yixuan Wang, Sachin Vaidya, Fabian Ruehle, James Halverson, Marin Soljačić, Thomas Y Hou, and Max Tegmark. 2024. Kan: Kolmogorov-arnold networks. arXiv preprint arXiv:2404.19756. — [arXiv:2404.19756](https://arxiv.org/abs/2404.19756) [find](https://scholar.google.com/scholar?q=Tsung%2DYi+Lin%2C+Priya+Goyal%2C+Ross+Girshick%2C+Kaiming+He%2C+and+Piotr+Doll%C3%A1r%2E+2017%2E+Focal+loss+for+dense+Ziming+Liu%2C+Yixuan+Wang%2C+Sachin+Vaidya%2C+Fabian+Ruehle%2C+James+Halverson%2C+Marin+Solja%C4%8Di%C4%87%2C+Thomas+Y+Hou%2C)
11. Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. In International Conference on Learning Representations. [find](https://scholar.google.com/scholar?q=Ilya+Loshchilov+and+Frank+Hutter%2E+2019%2E+Decoupled+weight+decay+regularization%2E+In+International+Conference+on+Learning+Representations%2E)
12. Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Vasiljevic, Pauline Sun, Peter Schwartz, Akhil Batra, Vincent Brandao, et al. 2024. [Dinov2: Learning robust visual features without supervision](https://md.research.elamur.ai/free/en/papers/oquab-2023-dinov2-learning-robust-visual-features/). Transactions on Machine Learning Research.
13. Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. 2019. [PyTorch: An imperative style, high-performance deep learning library](https://md.research.elamur.ai/free/en/papers/paszke-2019-pytorch-imperative-style-high-performance/). In Advances in Neural Information Processing Systems, volume 32.
14. Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. [Learning transferable visual models from natural language supervision](https://md.research.elamur.ai/free/en/papers/radford-2021-learning-transferable-visual-models-natural/). In International Conference on Machine Learning, pages 8748–8763. PM…
15. Mingxing Tan and Quoc Le. 2019. [Efficientnet: Rethinking model scaling for convolutional neural networks](https://md.research.elamur.ai/free/en/papers/tan-2019-efficientnet-rethinking-model-scaling-convolutional/). In International conference on machine learning, pages 6105–6114. PMLR.
16. Zhengzhong Tu, Hossein Talebi, Han Zhang, Feng Yang, Jian Sun, Laurent Schampers, Ang Li, Christian Szegedy, Caroline Pantofaru, Golnaz Ghiasi, et al. 2022. Maxvit: Multi-axis vision transformer. arXiv preprint arXiv:2204.01697. — [arXiv:2204.01697](https://arxiv.org/abs/2204.01697) [find](https://scholar.google.com/scholar?q=Zhengzhong+Tu%2C+Hossein+Talebi%2C+Han+Zhang%2C+Feng+Yang%2C+Jian+Sun%2C+Laurent+Schampers%2C+Ang+Li%2C+Christian+Szegedy%2C+Caroline+Pantofaru%2C+Golnaz+Ghiasi%2C+et+al%2E+2022%2E+Maxvit%3A+Multi%2Daxis+vision+transformer%2E+arXi)
17. Aäron van den Oord, Yazhe Li, and Oriol Vinyals. 2018. [Representation learning with contrastive predictive coding](https://md.research.elamur.ai/free/en/papers/oord-2018-representation-learning-contrastive-predictive-coding/). In arXiv preprint arXiv:1807.03748.
18. Ross Wightman. 2019. Pytorch image models. [https://github.com/rwightman/](https://github.com/rwightman/) pytorch-image-models. [↗](https://github.com/rwightman/)
19. Sangdoo Yun, Dongyoon Han, Seong Joon Oh, Sanghyuk Chun, Junsuk Choe, and Youngjoon Yoo. 2019. CutMix: Regularization strategy to train strong classifiers with localizable features. [find](https://scholar.google.com/scholar?q=Sangdoo+Yun%2C+Dongyoon+Han%2C+Seong+Joon+Oh%2C+Sanghyuk+Chun%2C+Junsuk+Choe%2C+and+Youngjoon+Yoo%2E+2019%2E+CutMix%3A+Regularization+strategy+to+train+strong+classifiers+with+localizable+features%2E)
20. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8024– 8033. [find](https://scholar.google.com/scholar?q=In+Proceedings+of+the+IEEE%2FCVF+International+Conference+on+Computer+Vision%2C+pages+8024%E2%80%93+8033%2E)
21. Hongyi Zhang, Moustapha Cai, Yann Dauphin, and David Lopez-Paz. 2018. [mixup: Beyond empirical risk minimization](https://md.research.elamur.ai/free/en/papers/zhang-2017-mixup-empirical-risk-minimization/). In International Conference on Learning Representations. 9.

*Updated 1 November 2026*