EUVIMEDEUVIMED entdecken
Crossref

Enhancing Visual Search Capabilities Through Visual Language Model

Harshavardhana Doddamani, Maulya Naik, Baba Fakruddin Ali B H, Prithviraj P, Nayana S, Rohith P

International Journal of Creative and Open Research in Engineering and Management · 13.06.2026 · 02 (06) · 1-9

DOI: 10.55041/ijcope.v2i6.162 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Creative and Open Research in Engineering and Management, online erschienen am 13.06.2026. Verfasst von Harshavardhana Doddamani und 5 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

General-purpose Vision-Language Models (VLMs) like CLIP are suffered from a significant "domain gap" when they are applied to specialized fields, failing to differentiate nuanced visual categories. While fine-tuning is a known solution, the critical, secondary "data noise problem" that is arisen from using LLMs for dataset creation is addressed by this paper. It was found that nearly 19% of our initial LLM-generated culinary dataset was consisted of generic, "noisy" captions (e.g., "A photo of a food dish"). This work presents a comprehensive end-to-end methodology anchored in a rigorous data refinement framework designed to eliminate noise. This is combined with an iterative, sequential fine-tuning strategy that progressively has the learning rate decayed to prevent overfitting. This combined method was proved highly effective, with the model's performance being transformed on unseen validation data from a 77.56% baseline (on noisy data) to a peak accuracy of 93.00% (on the refined dataset). A reproducible blueprint for adapting general VLMs to niche domains is provided by this work, demonstrating that methodical data refinement is considered as critical as the model's architecture.

Zitieren

APA

Doddamani, H., Naik, M., H, B. F. A. B., P, P., S, N., & P, R. (2026). Enhancing Visual Search Capabilities Through Visual Language Model. International Journal of Creative and Open Research in Engineering and Management, 02(06), 1-9. https://doi.org/10.55041/ijcope.v2i6.162

Vancouver

Doddamani H, Naik M, H BFAB, P P, S N, P R. Enhancing Visual Search Capabilities Through Visual Language Model. International Journal of Creative and Open Research in Engineering and Management. 2026;02(06):1-9. doi:10.55041/ijcope.v2i6.162

BibTeX
@article{Doddamani2026,
  author = {Doddamani, Harshavardhana and Naik, Maulya and H, Baba Fakruddin Ali B and P, Prithviraj and S, Nayana and P, Rohith},
  title = {Enhancing Visual Search Capabilities Through Visual Language Model},
  journal = {International Journal of Creative and Open Research in Engineering and Management},
  year = {2026},
  volume = {02},
  number = {06},
  pages = {1--9},
  doi = {10.55041/ijcope.v2i6.162}
}
Kostenlos starten →