BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//talks.osgeo.org//foss4g-uk-2026//talk//7FCBAN
BEGIN:VTIMEZONE
TZID:GMT
BEGIN:STANDARD
DTSTART:20001029T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:GMT
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T020000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:BST
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-foss4g-uk-2026-7FCBAN@talks.osgeo.org
DTSTART;TZID=GMT:20261012T150000
DTEND;TZID=GMT:20261012T153000
DESCRIPTION:Applying deep learning to large-scale geospatial imagery presen
 ts a fundamental trade-off: high-resolution datasets are simply too large 
 to analyse in a single pass. While downsampling preserves broader spatial 
 context\, it sacrifices fine-grained detail. Conversely\, slicing imagery 
 into discrete tile grids introduces arbitrary boundaries that bisect objec
 ts and isolate features from their surroundings. Consequently\, downstream
  detection and segmentation models receive truncated context\, compromisin
 g inference quality. While overlapping tiles offer a partial mitigation\, 
 they merely shift the burden to complex post-processing required to resolv
 e conflicting predictions.\n\nIn this talk\, we present a novel solution t
 o bridge tile boundaries using the latest vision foundation models like DI
 NOv3. DINOv3 processes raw imagery into uniform token grids\, where each p
 atch embedding captures rich\, high-dimensional spatial context without re
 lying on human annotations. By comparing these embeddings\, semantically a
 nd spatially coherent patches can be clustered across an image. We demonst
 rate how this unsupervised clustering capability extends beyond individual
  tile boundaries to encompass arbitrarily large regions. This approach red
 uces reliance on tile overlaps and lossy downsampling\, equipping downstre
 am detection and segmentation models with the unbroken context needed to a
 ccurately delineate complex geographical features.\n\nTo highlight its pra
 ctical value within geospatial workflows\, we demonstrate our Pytorch pipe
 line and showcase its performance against a diverse range of satellite and
  aerial imagery datasets.
DTSTAMP:20260821T121544Z
LOCATION:Create 2
SUMMARY:Thinking Outside the Box: Bridging Tile Boundaries in Large-Scale G
 eospatial Image Analysis - Madeleine Darbyshire
URL:https://talks.osgeo.org/foss4g-uk-2026/talk/7FCBAN/
END:VEVENT
END:VCALENDAR