Skip to content

4_DeepGlobe

dbuscombe-usgs edited this page Feb 23, 2023 · 1 revision

💾 DeepGlobe

DeepGlobe is a dataset for generic landuse/cover segmentation of high-resolution orthomosaic/satellite imagery

Classes:

  1. urban
  2. agricultural
  3. rangeland
  4. forest
  5. water
  6. bare
  7. unknown

Example labeled imagery

6399_sat_overlay 10452_sat_overlay 15573_sat_overlay 28935_sat_overlay 33573_sat_overlay

512x512 imagery

Training curve for the best model:

deepglobe_naipRGB_512_v6_trainhist_24

Some example validation outputs:

deepglobe_naipRGB_512_v6_val_324 deepglobe_naipRGB_512_v6_val_288 deepglobe_naipRGB_512_v6_val_294 deepglobe_naipRGB_512_v6_val_306 deepglobe_naipRGB_512_v6_val_309

Clone this wiki locally