BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pydata-amsterdam2026//speaker//QSVYUX
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-pydata-amsterdam2026-UPGM73@pretalx.com
DTSTART;TZID=CET:20260911T100500
DTEND;TZID=CET:20260911T105000
DESCRIPTION:Your model isn't the problem. Your data is.\n\nIn production co
 mputer vision\, teams consistently reach for model-level fixes\, bigger ba
 ckbones\, longer training runs\, more data\, while the underlying dataset 
 stays messy\, mislabelled\, and unrepresentative of the real world. We did
  the same\, until we stopped.\n\nThis talk presents three data strategies 
 we applied to a single\, deceptively simple problem: counting crates movin
 g through a logistics environment. The problem looked straightforward. The
  data was not.\n\nThe three strategies are:\n\n1. **Annotation quality at 
 scale**: using model-assisted labelling and embedding-based clustering to 
 systematically surface and correct errors hiding in production datasets  \
 n2. **Synthetic data generation**: covering rare but critical edge cases a
 bsent from real captures\, and navigating the domain gap between synthetic
  and real images  \n3. **Intelligent dataset reduction**: selecting maxima
 lly informative training subsets\, based on the counterintuitive principle
  that more data is not always better data\n\nThis is not a benchmarking ta
 lk. There are no leaderboard numbers here. The goal is to share a way of t
 hinking about data that you can apply to your own problems\, and to be hon
 est about where each strategy works\, where it struggles\, and how they in
 teract.\n\n**Target audience:** Data scientists\, ML engineers\, and appli
 ed researchers working with real-world computer vision or ML pipelines.\n\
 n**Takeaways:** A practical\, transferable framework for approaching data 
 quality.
DTSTAMP:20260710T150511Z
LOCATION:Main stage
SUMMARY:Data First\, Model Second: Three Strategies for Production Computer
  Vision - Alexander Kern\, Guus van der Ham
URL:https://pretalx.com/pydata-amsterdam2026/talk/UPGM73/
END:VEVENT
END:VCALENDAR