# PhD Topics Pyry/Guillaume

## Map Generalisation

To enable smooth zooming during the exploration of multi-scale maps, it is necessary to have maps progressively generalised and this amount of generalisation (21 zoom levels, possible maps even between the zoom levels) is only possible if the process is fully automated.

We want to develop fully automated multi-scale generalisations for both French and Finnish maps, with open source tools.

To generalise multi-scale maps, is it better to use a star, a ladder, or a mixed approach?

One of the challenges will be the consistency and the progressiveness of the map across scales.

Another challenge will be to handle large datasets.

## Deep Learning with Vector Data

Pattern recognition in maps really benefits from the structure of the vector map data with its topology and spatial relations, but current deep learning based techniques (e.g. Touya & Lokhat, 2020) only use images of the vector data to classify or segment the map data.

There are a few attempts to use deep convolutions on the vector data ([van't Veer et al., 2018](https://arxiv.org/abs/1806.03857); [Yan et al., 2019](http://www.sciencedirect.com/science/article/pii/S0924271619300437); [Yan et al., 2020](https://doi.org/10.1080/13658816.2020.1768260)), but we believe it is a promising way to analyse multi-scale vector map data.

First use cases we want to address: ??

## Eye Tracking and Multi-Scale Maps

A fluid anchor-based exploration of multi-scale maps, as proposed in the LostInZoom project, assumes that anchors are visually saliant in the map and during the zoom. Eye tracking techniques could be useful to measure what map readers look at during a multi-scale map exploration.

Eye trackers are often used in cartography, but the challenge here is that multi-scale maps are changing during the interactions, and it is not obvious to design an experimental setup that properly records the visual interactions with a multi-scale map.