Creating an ML-Export Tool

User Story

A user would like to perform machine learning against an area. They provide an area and the service responds with

1. Available machine learning features for that area: i.e.
    * Building Count
    * Building Area
    * Building Shapes
2. Estimated time to produce requested output
3. Output formats for result.  

Export End Points

1. TMS
2. Vector Tiles
3. GeoJson
4. Cloud Optimized GeoTiff

Test items:

Test Location 1: SpaceNet Area of Interest Las Vegas

Demo Map:

https://cwnets-demo.netlify.com/#/url/s3%3A%2F%2Fspacenet-dataset%2FAOI_2_Vegas%2FsrcData%2FrasterData%2FAOI_2_Vegas_MUL-PanSharpen_Cloud.tif/center/-115.24685,36.19648/zoom/17

srcImagery

* s3://spacenet-dataset/AOI_2_Vegas/srcData/rasterData/AOI_2_Vegas_MUL-PanSharpen_Cloud.tif
* s3://spacenet-dataset/AOI_2_Vegas/srcData/rasterData/AOI_2_Vegas_PAN_Cloud.tif
* s3://spacenet-dataset/AOI_2_Vegas/srcData/rasterData/AOI_2_Vegas_MUL_Cloud.tif 

results COG:

* s3://spacenet-dataset/AOI_2_Vegas/resultData/AOI_2_Vegas_MULPS_v13_cloud.tiff

results GeoJSON:

* s3://spacenet-dataset/AOI_2_Vegas/resultData/AOI_2_Vegas_MULPS_v13.geojson

results MBtiles:

* s3://spacenet-dataset/AOI_2_Vegas/resultData/AOI_2_Vegas_MULPS_v13.mbtiles

Example to pull tiles from COG GeoTIFF

from rio_tiler import main
import mercantile

address = "s3://spacenet-dataset/AOI_2_Vegas/resultData/AOI_2_Vegas_MULPS_v13_cloud.tiff"


# -115.24, 36.1, -115.2, 36.2, 17)
tile_coords = mercantile.tile(-115.24, 36.1986, 18)

tile, mask = main.tile(address,
                           tile_coords.x,
                           tile_coords.y,
                           tile_coords.z
                           )

Result Aggregator:

1. Create ML-Tile List = 12*z19 Tile (1024pix x 1024 pix tiles) with 1 z19 tile overlap (25% overlap)
2. Pass new ML-Tile to ML-Algorithm:
3. Recieve ML-Algorithm output: 
    
    * PNG:    
        * Band 1: RasterTile_Building_Center
        * Band 2: RasterTile_Building_Edge (Optional)
    * Threshold_Point: (0,1)
    * VectorTile: (Vector Tile)
    
4. Combine ML-Algorithm output:
    * Cloud Optimized GeoTiff
        Band 1: RasterTile_Building_Center
        Band 2: RasterTile_Building_Edge (Optional)
        GEOTIFF_TAG: Threshold_Value
    * GeoJSON
    * VectorTile: Z18
    * RasterTile: Z18

ML-Algorithm:

## input: 
* List of JSON: 
    [{'imageId': imageId,
     'imageLoc': https://bucket-in/z/x/y.png,
     'dst_raster':   https://bucket-out/z/x/y.png,
     'dst_vector':   {https://bucket-out/z/x/y.json,
     }] 

image tile for imageLoc should return a:

    12 * z19 resolution (z17 scope): 
    
    |x-2,y-2|x-1,y-2|x+1,y-2|x+2,y-2|
    |x-2,y-1|x-1,y-1|x+1,y-1|x+2,y-1|
    |x-2,y+1|x-1,y+1|x+1,y+1|x+2,y+1|  = (1024x1024 pixel 3-Band png 8 bit jpeg) = Z18 x 4x SR Tile
    |x-2,y+2|x-1,y+2|x+1,y+1|x+2,y+2|

## output: 
* 1024x1024  x 2 bands 

    Band 1: RasterTile_Building_Center
    Band 2: RasterTile_Building_Edge (Optional)
    GEOTIFF_TAG: Threshold_Value

* 1024 x 1024 Vector Tile

* geojson: Pixel Coordinates

Key pieces of information for model to report

1. Estimated Time for Production {x seconds/tile}
2. Accuracy {Precision/Recall, Mean Error)
3. Resolution Required {zoomTile, GSD}
4. Output Resolution {zoomTile, GSD}
5. Features {'Building Count', 'Building Area', 'Building Bounding Box', 'Building Instance Segmentation'}
6. Region Trained: {STAC_Item}
7. Test Data:
8. src docker container

Type of Requests (Tasking Manager to Service):

Get Requests:

* endpoint/getmodels
    * report registered models


* endpoint/getmodelcapability/{modelname}
    * report key pieces above
    
* endpoint/estimate/{modelname, geobox, datasource}
    * report estimated time of completion (Days/Hours/Minutes)
    * report estimated cost
    
* endpoint/status/{processid}
    * report process status
    * report estimated completion
    
* endpoint/orderexport/orderid/tms/{x}/{y}/{z}.png
* endpoint/orderexport/orderid/vectortile/{x}/{y}/{z}.pbf

Post Requests:

* endpoint/orderexport/{modelversion, geobox, datasource}
    * return orderId for status

Type of Request (Service to ML Model)

Get Requests:

* endpoint/{modelid}/{x}/{y}/{z}.pbf
    * return vector tile for specified model request
    
* endpoint/{modelid}/{x}/{y}/{z}.png
    * return raster tile for specified model request

Post Requests:

 * endpoint/{modeljson}}
    * register new model