--- title: "AyahVerse at NakbaArchiveClassifier Shared Task: Architectural" description: "This paper describes the AyahVerse team's entry to the NakbaArchiveClassifier shared task at Nakba-NLP 2026. 4 references." authors: ["Ibad-ur-Rehman Rashid", "Akhtar Ali"] year: 2026 venue: "Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026" source: "https://aclanthology.org/2026.nakbanlp-1.30/" doi: "10.63317/3p7n5t5hmz6h" 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/rashid-2026-ayahverse-nakbaarchiveclassifier-shared-task-architectural/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/rashid-2026-ayahverse-nakbaarchiveclassifier-shared-task-architectural/) [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/23/) / AyahVerse at NakbaArchiveClassifier Shared Task: Architectural Trade-offs and Decision Calibration for Humanitarian Image Classification # AyahVerse at NakbaArchiveClassifier Shared Task: Architectural Trade-offs and Decision Calibration for Humanitarian Image Classification **Authors:** [Ibad-ur-Rehman Rashid](https://md.research.elamur.ai/free/en/authors/ibad-ur-rehman-rashid/) and Akhtar Ali · **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-1.30/) · [PDF](https://aclanthology.org/2026.nakbanlp-1.30.pdf) · [DOI 10.63317/3p7n5t5hmz6h](https://doi.org/10.63317/3p7n5t5hmz6h) · [OpenAlex](https://openalex.org/W7166585932) **Citations:** 0 citations (OpenAlex, 2026-10-14) ## Summary This paper describes the AyahVerse team's entry to the NakbaArchiveClassifier shared task at Nakba-NLP 2026. The task is to label Instagram images from Gaza as showing either destroyed or intact infrastructure. Humanitarian deployments often lack strong GPUs, so the authors compare a heavy CNN (ResNet50) with two compact ones (MobileNetV2 and EfficientNet-B0). All three start from ImageNet weights and use identical training settings. Intact scenes outnumber destroyed ones, so training uses Focal Loss. After training, a grid search on validation data selects the sigmoid cut-off that maximizes Macro F1. The authors also test whether larger input images help ResNet50. EfficientNet-B0 achieved the best blind-test Macro F1 (0.85), with MobileNetV2 close behind. Both beat ResNet50 despite having far fewer parameters. Higher resolution and a lowered threshold each improved ResNet50, but neither lifted it above the compact models. The work matters as practical evidence that inexpensive models can suffice for damage triage in crisis image archives. However, the conclusions rest on one modest dataset and results reported without variance estimates. ## Key points - **Dataset:** 2,001 Instagram images collected October 2023 to December 2025. Of these, 706 (35.3%) show destruction and 1,295 (64.7%) are intact. The official train and validation sets (1,599 images) were merged and re-split 80/20 with stratification into 1,279 training and 320 validation images. - **Models and training:** ResNet50 (~25.5M parameters), MobileNetV2 (~3.4M) and EfficientNet-B0 (~5.3M). Training used Focal Loss (α=1, γ=2) and AdamW (lr 1e-4) with ReduceLROnPlateau (patience 3) on a single T4 GPU. Inputs were 224×224, plus one ResNet50 variant at 334×334. - **Test Macro F1:** - EfficientNet-B0: 0.85 - MobileNetV2: 0.84 - ResNet50 at 334×334: 0.83 - ResNet50 with threshold 0.45: 0.81 - ResNet50 baseline: 0.79 - **Threshold calibration:** The search covered roughly 0.3–0.75. ResNet50's optimum was 0.45, which barely changed validation F1 (0.8197 → 0.8214) but raised test F1 from 0.79 to 0.81. Both compact models performed best at the default 0.50. The authors read this as larger models being more prone to majority-class bias, although only one heavy model was tested. - **Code:** The code is released on GitHub (Ebad-urRehman/nakba_archive_efficient_classification). *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* 📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 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 · [⚙️ Civil and Structural Engineering](https://md.research.elamur.ai/free/en/shelves/engineering/civil-and-structural-engineering/) · 2 works ## Cite Cite as: Ibad-ur-Rehman Rashid and Akhtar Ali. 2026. AyahVerse at NakbaArchiveClassifier Shared Task: Architectural Trade-offs and Decision Calibration for Humanitarian Image 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{rashid2026ayahverse, title = {{AyahVerse at NakbaArchiveClassifier Shared Task: Architectural Trade-offs and Decision Calibration for Humanitarian Image Classification}}, author = {Ibad-ur-Rehman Rashid and Akhtar Ali}, booktitle = {Proceedings of the 2nd International Workshop on Nakba Narratives as Language Resources @ LREC 2026}, year = {2026}, doi = {10.63317/3p7n5t5hmz6h}, url = {https://doi.org/10.63317/3p7n5t5hmz6h} } ``` - [← previous AyahVerse at AbjadGenEval Shared Task: Monolingual Precision and Cross-Lingual Analysis in Perso-Arabic AI Detection](https://md.research.elamur.ai/free/en/papers/nawaz-2026-ayahverse-abjadgeneval-shared-task-monolingual/) - [next → AYN: A Tiny Yet Competitive Indian Legal Language Model Pretrained from Scratch](https://md.research.elamur.ai/free/en/papers/niyogi-2026-ayn-tiny-yet-competitive-indian/) - [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/23/) ## References 4 references, 2 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](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) 2. Firoj Alam, Ferda Ofli, and Muhammad Imran. 2018. Crisismmd: Multimodal twitter datasets from natural disasters. [find](https://scholar.google.com/scholar?q=Firoj+Alam%2C+Ferda+Ofli%2C+and+Muhammad+Imran%2E+2018%2E+Crisismmd%3A+Multimodal+twitter+datasets+from+natural+disasters%2E) 3. Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2015. [Deep residual learning for image recognition](https://md.research.elamur.ai/free/en/papers/he-2015-deep-residual-learning-image-recognition/). 4. Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. 2018. [Focal loss for dense object detection](https://md.research.elamur.ai/free/en/papers/lin-2017-focal-loss-dense-object-detection/). \#\# Conclusion *Updated 1 November 2026*