developmentseed / skynet-data

[DEPRECATED] Data pipeline for machine learning with OpenStreetMap
ISC License
169 stars 32 forks source link

Use TRAIN_TILES variable as expected #12

Open ellieLitwack opened 7 years ago

ellieLitwack commented 7 years ago

Currently, TRAIN_TILES sets the size of sample.txt. This results in unexpected and undesirable behavior. After running remove-bad-images and setting a LABEL_RATIO greater than zero, many or most images are filtered from some datasets. Thus, even though the user may set TRAIN_TILES to 1000, they can end up with only a hundred images. TRAIN_TILES should set the size of train.txt - samples.txt should be every tile at the user's set ZOOM_LEVEL.