2_Chesapeake

Jump to bottom

Daniel Buscombe edited this page Feb 25, 2023 · 2 revisions

💾 Chesapeake Landcover

Chesapeake is a dataset for generic/coastal landuse/cover mapping on high-resolution orthomosaic imagery

Classes:

water
tree_canopy_forest
low_vegetation_field
barren land
impervious_other
impervious_road
no_data

Example labeled imagery

m_3607632_ne_18_1_naip-old-2_overlay m_3707549_ne_18_1_naip-old-1_overlay m_3707602_sw_18_1_naip-old-0_overlay m_3707626_nw_18_1_naip-old-2_overlay

Model for 7 classes, 512x512 imagery

Zenodo release: https://zenodo.org/record/7576904#.Y-FnkRzMLRY

Training curve for the best model:

ches_7class_naipRGB_512_v5_trainhist_24

Some example validation outputs:

ches_7class_naipRGB_512_v5_val_367

Segformer model for 7 classes, 512x512 imagery

Zenodo release: https://zenodo.org/record/7677506#.Y_pdBIDMJkg
segmentation zoo model name: chesapeake_7class_segformer_7677506
Buscombe, Daniel. (2023). Doodleverse/Segmentation Zoo/Seg2Map Segformer models for Chesapeake/7-class segmentation of RGB 512x512 high-res. images (v1.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.7677506

Training curve for the best model:

Some example validation outputs:

ches_7class_naipRGB_512_segformer_v3_val_245