Skip to content

Generate landslide detection training labels#43

Draft
APatrickJ wants to merge 8 commits intomainfrom
patrickj/landslides
Draft

Generate landslide detection training labels#43
APatrickJ wants to merge 8 commits intomainfrom
patrickj/landslides

Conversation

@APatrickJ
Copy link
Copy Markdown
Contributor

@APatrickJ APatrickJ commented Jan 6, 2026

  • Create positive samples (landslides) and negative samples as ring buffers around landslide polygons
    • Negative rings are created with a gap from the positive and a specified ring width
    • Both use the same event_date (no temporal offset)
  • Remove overlapping portions from negative samples
    • Positives are buffered by gap_width before removal to match the ring creation approach
    • Only positives that occurred on or before the negative's time are considered
  • Balance dataset geographically and by label
  • Compute bounding box metrics and task geometries using per-geometry UTM/UPS CRS

Example (red = positive landslide samples, blue = negative no landslide samples):

landslide-sample

@APatrickJ APatrickJ self-assigned this Jan 6, 2026
@APatrickJ APatrickJ changed the title Generate landslide detection training labels from inventory data Generate landslide detection training labels Jan 8, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant