I had thought that using
BoundingBox() would cause clipping, but this code:
r = TVF2.convert_format_bounding_box( datapoints.BoundingBox(ltwh, format=datapoints.BoundingBoxFormat.XYWH, spatial_size=img_hw, ), new_format=datapoints.BoundingBoxFormat.XYXY, )
xyxy regardless of the spatial size I give it. Does it only store it for usage down the pipeline?
Also, I see in the documentation that
(h, w) whereas
XYWH implies the reverse order, why is this not consistent? perhaps I’m missing something?