Register
Login
Resources
Docs Blog Datasets Glossary Case Studies Tutorials & Webinars
Product
Data Engine LLMs Platform Enterprise
Pricing Explore
Connect to our Discord channel
Christian Werner 1d5c2316d1
Fix cli model handling for inference script and proper channel matching for ensemble inference in case of RGB data
2 years ago
..
e849e87c3a
Update readme (#62)
2 years ago
e6a0c29265
Consolidate data handling util routines
2 years ago
c7a25e50c9
Rerun inference for all 4 years, update the normalization settings in the data section and cleanup inference stats code
2 years ago
9fc09fed52
Reverse src layout to original deadtrees source folder
2 years ago
e9431a521b
Disable fp16 training and consequenctly reduce default batch_size and switch to dice loss to fix instability
2 years ago
19f865aa19
Minor update to preprocessign scripts
2 years ago
1d5c2316d1
Fix cli model handling for inference script and proper channel matching for ensemble inference in case of RGB data
2 years ago
34980bccb0
Fix wrong folder in mergedatasets
2 years ago

README.MD

You have to be logged in to leave a comment. Sign In

Preprocessing Pipeline 🪠

Stage 0 (not tracked)

Data origin

Data is sources from the The luxembourgish data platform.

Data is provided as JPEG2000 encoded 3-band images with 20cm resolution and projected in EPSG:2169 (LUREF). Data for 2020 is provided in 10cm resolution and downsampled in preprocessing to match the other years.

Initial recombination and reprojection

Prior to the DVC-tracked preprocessing steps the raw data was reorganized as follows (the final output files are tracked in this repo though):

YEAR=2020

if [[ $YEAR -eq 2017 ]]
then
  RGB=ortho2017_RVB_pays.jp2
  NIR=LUX17_CIR_v3.jp2
elif [[ $VAR -eq 2018 ]]
then
  RGB=ortho2018_RGB_pays.jp2
  NIR=ortho2018_CIR_pays.jp2
elif [[ $VAR -eq 2019 ]]
then
  RGB=ortho_2019.jp2
  NIR=ortho_2019ir.jp2
elif [[ $VAR -eq 2020 ]]
then
  RGB=Luxembourg-2020_ortho10cm_RVB_LUREF.jp2
  NIR=Luxembourg-2020_ortho10cm_PIR_LUREF.jp2
else 
  echo "Undefined Year: ${YEAR}"
  exit -1
fi


# merge bands (allow projection difference since one source has a differnt projection string but correct projection parameters)
gdalbuildvrt b1.vrt -b 1 ${RGB} 
gdalbuildvrt b2.vrt -b 2 ${RGB}
gdalbuildvrt b3.vrt -b 3 ${RGB}
gdalbuildvrt b4.vrt -b 1 ${NIR}
gdalbuildvrt -allow_projection_difference -separate allbands.vrt b1.vrt b2.vrt b3.vrt b4.vrt

gdal_translate -of GTiff -co "COMPRESS=LZW" -co "PREDICTOR=2" -co "TILED=YES" -co "BIGTIFF=YES" -co "ALPHA=NO" -colorinterp_1 "red" -colorinterp_2 "green" -colorinterp_3 "blue" allbands.vrt ortho_ms_${YEAR}.tif

# reproject and limit extent to common area
gdalwarp -co "NUM_THREADS=ALL_CPUS" -multi  -overwrite -ot Byte -ts 302346 420442 -te 263584.672 5479988.018 324060.605 5564085.859 -t_srs '+proj=utm +zone=32 +ellps=GRS80 +towgs84=0,0,0,0,0,0,0 +units=m +no_defs' -r average ortho_ms_2017.tif ortho_ms_${YEAR}_EPSG3044.vrt
gdal_translate -co "NUM_THREADS=ALL_CPUS" -co "COMPRESS=LZW" -co "PREDICTOR=2" -co "TILED=YES" -co "BIGTIFF=YES" -co "ALPHA=NO" ortho_ms_${YEAR}_EPSG3044.vrt ortho_ms_${YEAR}_EPSG3044.tif

Tip!

Press p or to see the previous file or, n or to see the next file

Comments

Loading...