lewfish

this is a fake CLA for testing purposes

lewfish

./output/
├── bundle
│   └── model-bundle.zip
├── eval
│   └── eval.json
├── Makefile
├── pipeline-config.json
├── predict
│   └── 6_12
│       └── labels.tif

lewfish

backend_config = PyTorchSemanticSegmentationConfig(
    data=data_config,
    model=SemanticSegmentationModelConfig(
        external_def=ExternalModuleConfig(
            github_repo='AdeelH/pytorch-fpn:0.2',
            name='fpn',
            entrypoint='make_fpn_resnet',
            entrypoint_kwargs={
                'name': 'resnet50',
                'fpn_type': 'panoptic',

lewfish

aug_transform = A.Compose([
    A.Flip(),
    A.ShiftScaleRotate(),
    A.OneOf([
        A.RGBShift(),
        A.ToGray(),
        A.ToSepia(),
    ]),
    A.OneOf([
        A.RandomBrightness(),

lewfish

train_scene = SceneConfig(
    id='2_10',
    raster_source=RasterioSourceConfig(uris=['./2_10_RGBIR.tif']),
    label_source=SemanticSegmentationLabelSourceConfig(
        rgb_class_config=class_config,
        raster_source=RasterioSourceConfig(uris=['./2_10_label.tif'])),
    aoi_uris=['./2_10_aoi.json'])

lewfish

data_config = SemanticSegmentationGeoDataConfig(
    scene_dataset=dataset_config,
    # read 200 randomly located chips of size 300x300 from the scene
    window_opts=GeoDataWindowConfig(
        method=GeoDataWindowMethod.random,
        size=300,
        size_lims=(200, 400),
        max_windows=200))

lewfish

train_scene = SceneConfig(
    id='2_10',
    raster_source=RasterioSourceConfig(
        # include all 4 bands in the channel_order
        uris=['./2_10_RGBIR.tif'], channel_order=[0, 1, 2, 3]),
    label_source=SemanticSegmentationLabelSourceConfig(
        rgb_class_config=class_config,
        raster_source=RasterioSourceConfig(uris=['./2_10_label.tif'])))
...

lewfish

val_scene = SceneConfig(
    label_store=SemanticSegmentationLabelStoreConfig(
        rgb=True,
        vector_output=[
            PolygonVectorOutputConfig(class_id=i)
            for i in range(len(class_config))
        ]),
    smooth_output=True,
    smooth_as_uint8=True)

lewfish

from rastervision.pipeline.runner import *
from rastervision.core.rv_pipeline import *
from rastervision.core.data import *
from rastervision.pytorch_backend import *
from rastervision.pytorch_learner import *


def get_config(runner: Runner, root_uri: str) -> SemanticSegmentationConfig:
    # specify class names and corresponding colors in the label rasters
    class_config = ClassConfig(

lewfish

import boto3
from pystac import STAC_IO
from pystac import Catalog, LabelItem
from shapely.geometry import box

def setup_stac_s3():
    def my_read_method(uri):
        parsed = urlparse(uri)
        if parsed.scheme == 's3':
            bucket = parsed.netloc