Abstract
Video geolocalization is a crucial problem in current times. Given just a video, ascertaining where it was captured from can have a plethora of advantages. The problem of worldwide geolocalization has been tackled before, but only using the image modality. Its video counterpart remains relatively unexplored. Meanwhile, video geolocalization has also garnered some attention in the recent past, but the existing methods are all restricted to specific regions. This motivates us to explore the problem of video geolocalization at a global scale. Hence, we propose a novel problem of worldwide video geolocalization with the objective of hierarchically predicting the correct city, state/province, country, and continent, given a video. However, no large scale video datasets that have extensive worldwide coverage exist, to train models for solving this problem. To this end, we introduce a new dataset, "CityGuessr68k" comprising of 68,269 videos from 166 cities all over the world. We also propose a novel baseline approach to this problem, by designing a transformer-based architecture comprising of an elegant "Self-Cross Attention" module for incorporating scenes as well as a "TextLabel Alignment" strategy for distilling knowledge from textlabels in feature space. To further enhance our location prediction, we also utilize soft-scene labels. Finally we demonstrate the performance of our method on our new dataset as well as Mapillary(MSLS).
Method Overview
VideoMAE encoder outputs feature embeddings of the input video. The embeddings are then passed into 4 classifiers pertaining to 4 hierarchies. Their predictions are used for computing Geolocalization loss. Simultaneously prediction vectors are input into the Self-Cross Attention module, where vectors of all 4 hierarchies are concatenated and are attended to, by themselves and by each other to generate an intermediate attended vector(PV'). In the attention weights(w), the single colored weights along the diagonal refer to self attention weights, while the gradient double colored weights are the cross attention weights between vectors of those two different hierarchies. PV' is passed simultaneously through FFNs to generate vector PV's for Scene loss computation, and to the TextLabel Alignment module. There, it is passed through FFNt to generate vector PV't. PV't is used for TextLabel Alignment with feature embeddings Ft generated by the pretrained text-encoder from the label names of all 4 hierarchies.