Auto-dispatching wrapper that loads a registered object detector.
Resolves names like "faster_rcnn" / "detr_resnet50" /
"yolo_v3" to their concrete {Family}ForObjectDetection
subclass.
Notes
The detector forward returns an ObjectDetectionOutput with
logits (class scores per proposal / query) and pred_boxes
(predicted box coordinates). Two-stage detectors additionally
populate proposals so postprocessing can be reconstructed from
the output alone.
Examples
>>> from lucid.models import AutoModelForObjectDetection
>>> model = AutoModelForObjectDetection.from_pretrained("detr_resnet50")
>>> type(model).__name__
'DETRForObjectDetection'