AdFinder
Research-driven advertisement detection for printed media
AdFinder is Media Research AB’s platform for identifying and structuring advertisements in newspapers and other printed media. Its development is closely connected to the company’s research in artificial intelligence and computer vision, including the peer-reviewed studies AdVision and AdAPT.
AdVision investigated how deep-learning-based object detection can be used to automatically locate advertisements on newspaper pages. The study compared several families of detection models and evaluated both detection accuracy and computational efficiency across newspaper material from multiple countries and languages.
AdAPT builds on this work by addressing a different challenge: how advertisement-detection models can remain useful when the characteristics of the incoming material change. The research investigates domain adaptation and pseudo-labeling methods for adapting detection models to new newspaper domains while reducing the need for extensive manual annotation.
Together, AdVision and AdAPT contribute to the technical foundation and continued development of AdFinder. The research is carried out by Faeze Zakaryapour Sayyad, Industrial PhD researcher in Computer Engineering at Mid Sweden University and Media Research AB, together with academic collaborators.
Why this research matters
Newspaper layouts, advertisement formats and visual styles vary between publications, countries and over time. An advertisement may occupy an entire page or appear as a small element within a complex layout, making reliable automated detection a challenging computer vision problem.
Media Research works with heterogeneous media material at scale. Research on advertisement detection and model adaptation therefore addresses challenges that arise directly in the company’s work and contributes to the development and quality assurance of AdFinder and related media analytics workflows.
The publications below state Media Research AB, Ringvägen 31, 831 37 Östersund, Sweden as author affiliation. Media Research AB (legal name Media Research i Östersund AB, reg. no. 556662-4499) is a Swedish data company that collects visual advertising data across the Nordics and the UK via AdFinder and DailyAds.
Publications
AdVision: An efficient and effective deep learning based advertisement detector for printed media
Faeze Zakaryapour Sayyad (Media Research AB, Östersund · Mid Sweden University), Irida Shallari (Mid Sweden University), Seyed Jalaleddin Mousavirad (Mid Sweden University), Mattias O’Nils (Mid Sweden University), Faisal Z. Qureshi (University of Ontario Institute of Technology)
AdAPT: Advertisement detector adaptation under newspaper domain shift with null-based pseudo-labeling
Addresses the problem that a detector trained on one newspaper loses accuracy on another, using confidence-guided pseudo-labeling for unsupervised domain adaptation.
Model Evaluation and Selection for Robust and Efficient Advertisement Detection in Print Media
Compares models on mAP, F1 and inference latency for advertisement detection in print.
Funding
This work was funded in part by the Knowledge Foundation within the Industrial graduate school Smart Industry Sweden, together with Media Research AB.
Faeze Zakaryapour Sayyad
Industrial PhD researcher in Computer Engineering, Mid Sweden University · Media Research AB, Östersund
Faeze Zakaryapour is an Industrial PhD researcher in Computer Engineering at Mid Sweden University and is employed at Media Research AB. Her doctoral research is closely connected to the company’s AI development and focuses on artificial intelligence and computer vision, particularly object detection, document AI, and domain adaptation during distribution shift. So far, her research has resulted in publications including AdVision and AdAPT, whose methods have been successfully implemented in AdFinder, Media Research AB’s advertisement analysis platform. Her work is driven by real-world industry needs and contributes directly to Media Research AB’s products and workflows.
faeze.zakaryapoursayyad@miun.se
ORCID 0000-0001-6418-2876
Google Scholar · LinkedIn
Media Research AB, legal name Media Research i Östersund AB, reg. no. 556662-4499. Ringvägen 31, 831 37 Östersund, Sweden. mediaresearch.se