Detail of the student project

List
Topic:IM2GPS: image retrieval for geolocalization
Department:Katedra kybernetiky
Supervisor:Georgios Tolias, Ph.D.
Announce as:Diplomová práce, Semestrální projekt
Description:Given a large collection of geo-tagged images, e.g. street-view images, the goal is to build an image-based localization system which will predict the location at which an outdoor test image was taken.

The task will be handled as image retrieval where a large collection of images needs to be indexed based on visual representation extracted with deep learning models.Given a test image, the top retrieved images from the collection, along with their GPS-labels are used to predict the location of the test image. The goal is to exploit transformer architectures to facilitate such prediction.
Bibliography:G. Kordopatis-Zilos et al. ICMR 2021, Leveraging EfficientNet and Contrastive Learning for Accurate Global-scale Location Estimation
Vo Jacobs Hays, ICCV 2017, Revisiting IM2GPS in the Deep Learning Era
T. Wayend et al. ECCV 2016, Planet-photo geolocation with convolutional neural networks
Radenovic Tolias Chum, PAMI 2019, Fine-tuning CNN Image Retrieval with No Human Annotation

https://towardsdatascience.com/illustrated-guide-to-transformers-step-by-step-explanation-f74876522bc0
Responsible person: Petr Pošík