ANC Workshop - 28th January 2025 Title: Evaluating Monocular Depth Perception in Large Vision Models Speaker: Duolikun Danier, Postdoc ELIAI, University of Edinburgh, https://danier97.github.io/Abstract: Large-scale pre-trained vision models are becoming increasingly prevalent, offering expressive and generalizable visual representations. Recent studies have revealed their high-level geometric understanding, in particular in the context of depth perception. However, it remains unclear how depth perception arises in these models when explicit depth supervision is not provided during pre-training. In this talk, I will present recent work where we examine whether the monocular depth cues, similar to those used by the human visual system, emerge in these models.Title: Global-scale Species Distribution ModellingSpeaker: Oisin Mac Aodha, University of Edinburgh, https://homepages.inf.ed.ac.uk/omacaod/Abstract: Knowing the precise geographic locations where all species on earth can be found is a key piece of information for global-scale biodiversity monitoring. It underlies our ability to estimate the threatened status of different species and for prioritizing locations for conservation intervention. Citizen science efforts are producing large quantities of data that can be used to train models to estimate species' spatial distributions, with approximately three billion observations collected and publicly shared to date. However, this data exhibits significant biases, e.g. it is spatially, temporally, and taxonomically biased. In this talk, I will discuss recent work from my group, and with collaborators, where we are developing new machine learning solutions to address these issues. Jan 28 2025 13.00 - 14.00 ANC Workshop - 28th January 2025 Event organised by Oisin Mac Aodha. G.03, Informatics Forum
ANC Workshop - 28th January 2025 Title: Evaluating Monocular Depth Perception in Large Vision Models Speaker: Duolikun Danier, Postdoc ELIAI, University of Edinburgh, https://danier97.github.io/Abstract: Large-scale pre-trained vision models are becoming increasingly prevalent, offering expressive and generalizable visual representations. Recent studies have revealed their high-level geometric understanding, in particular in the context of depth perception. However, it remains unclear how depth perception arises in these models when explicit depth supervision is not provided during pre-training. In this talk, I will present recent work where we examine whether the monocular depth cues, similar to those used by the human visual system, emerge in these models.Title: Global-scale Species Distribution ModellingSpeaker: Oisin Mac Aodha, University of Edinburgh, https://homepages.inf.ed.ac.uk/omacaod/Abstract: Knowing the precise geographic locations where all species on earth can be found is a key piece of information for global-scale biodiversity monitoring. It underlies our ability to estimate the threatened status of different species and for prioritizing locations for conservation intervention. Citizen science efforts are producing large quantities of data that can be used to train models to estimate species' spatial distributions, with approximately three billion observations collected and publicly shared to date. However, this data exhibits significant biases, e.g. it is spatially, temporally, and taxonomically biased. In this talk, I will discuss recent work from my group, and with collaborators, where we are developing new machine learning solutions to address these issues. Jan 28 2025 13.00 - 14.00 ANC Workshop - 28th January 2025 Event organised by Oisin Mac Aodha. G.03, Informatics Forum
