Turning off windowed read?

[fangzp]

Thanks for building such a helpful tool for converting georeferenced vector annotations to COCO-compatible json.

I'm trying to generate COCO-formatted labels for agricultural fields in the AI4Boundaries dataset, which includes vector annotations for Sentinel-2 image tiles as well as an aerial orthophoto centered in but smaller (by a factor of 1/4th) than the S2 tiles. Ultimately I would like to have the aerial orthophoto as the reference image to be passed into my DL model with the COCO-formatted labels.

When I tried generating the COCO-formatted labels using the S2 images I noticed that the annotations had been split up into 4 parts, which is not the behavior I expected. I observed that the append_dataset function requires a window_bounds which, I think, needs to be smaller than the size of the image. At least when I try to pass a window_bounds=[(512,512)] into the append_dataset I get an error saying that no WindowSchema objects could be created from the given bounds, because the width_step and height_step which are passed into WindowSchema need to be greater than 0. Is there a way to turn off the windowed read entirely, since I would rather not use it for my conversion of the labels?

Relevant code:

test_img = # <-- path to orthophoto TIFF
test_label = # <-- path to geojson
output_dir = # <-- pathlib to output_dir
output_name = # <-- pathlib filename to give output

description = 'Test AI4B dataset'
contributor = 'fangzp'

dataset = create_dataset(description=description, contributor=contributor)
with open(test_label) as f:
    labels = geojson.load(f)
raster_source = rasterio.open(test_img)

labels = gpd.GeoDataFrame.from_features(labels['features'], crs=labels['crs']['properties']['name'])

dataset = append_dataset(dataset=dataset,
                         images_dir = output_dir, # output directory for all label images
                         labels=labels, # GeoDataFrame containing labels and class info
                         src=raster_source, # open rasterio reader for input raster
                         window_bounds=[(raster_source.width, raster_source.height)], # list of window bounds for each tile
                         id_attribute='category_id', #Column containing category_id values
                         name_attribute='category_name', # Column with category_name values
                         super_attribute=None) # Optional column with supercategory values

Final function is where I'm getting the WindowSchema error. Thanks in advance!