Create size benchmark with tif compression - Githubissues

daniel-j-h / nedem

Neural implicit digital elevation model

MIT License

2 stars 0 forks source link

Create size benchmark with tif compression #4

Open daniel-j-h opened 2 years ago

daniel-j-h commented 2 years ago

We should look into

[ ] compression options for rio/gdal to get the size of our quadkey down
[ ] quantize e.g. to int and then use rio/gdal compression

The goal should be to have a table of

quantization/compression technique
error metrics compared to ground truth

Once we have that, we can grid-search over model configurations trying to beat the quantization/compression from rio/gdal.

Read

Check rio/gdal output options

-co compress=deflate -co predictor=2