feat: Add satellite imagery classification job bundle#231
Conversation
github-actions
Bot
added
the
waiting-on-maintainers
| deadline bundle submit job_bundles/satellite_classification/ \ | ||
| -p TilesDir=/path/to/my/tiles |
There was a problem hiding this comment.
We should be clear that they'll have to modify the job bundle template to support this. Right now the file names are hard-coded in template.yaml.
| from matplotlib.patches import Patch | ||
|
|
||
|
|
||
| CLASS_NAMES = {1: "Water", 2: "Vegetation", 3: "Bare Soil", 4: "Rock", 5: "Cloud/Snow"} |
There was a problem hiding this comment.
these constants are in both classify_tile.py and mosaic.py - can we consolidate them somewhere?
| - name: attr.worker.os.family | ||
| anyOf: ["linux"] |
There was a problem hiding this comment.
maybe worth mentioning that this only works on linux fleets in the README?
There was a problem hiding this comment.
Removing this altogether. Upon further research, both linux and windows are supported
| TILE_CONFIGS = [ | ||
| { | ||
| "name": "T12S_GC_North_Rim", | ||
| "bounds": (-112.2, 36.2, -112.0, 36.3), | ||
| "description": "North Rim - dense vegetation + exposed rock", | ||
| "mix": {"vegetation": 0.45, "rock": 0.40, "soil": 0.10, "water": 0.0, "cloud": 0.05}, | ||
| }, |
There was a problem hiding this comment.
Where did these tiles come from? Do we need an attribution for them? Better to add a reference to their source even if not required so someone can see where they came from.
There was a problem hiding this comment.
All tiles are procedurally generated. Adding a docstring to generate_sample_tiles.py to detail this.
Demonstrates embarrassingly parallel tile processing on Deadline Cloud. Classifies synthetic Sentinel-2 tiles into 5 land-cover classes using spectral index thresholds, then merges results into a mosaic. Signed-off-by: Phillip Krasnick <259470369+phil-IO-p@users.noreply.github.com>
phil-IO-p
force-pushed
the
feat/satellite-classification
branch
from
0674f89 to
77c5be4
Compare
3 participants
Summary
Test plan
openjd checkpasses on the templatedeadline bundle submit job_bundles/satellite_classification/on a farm and verified output