efficientdet_d0(pretrained: bool = False, overrides: object = {})EfficientDet-D0 ().
Builds the smallest compound-scaled EfficientDet variant: EfficientNet-B0 backbone, BiFPN with 64 channels and 3 repeats, head depth 3, and a 512x512 input resolution. Approximately 3.9M parameters; COCO test-dev mAP of 33.8% (paper Table 2) at 2.5 BFLOPs.
Model Size
Parameters
pretrainedbool= FalseReserved for future pretrained-weight loading. Currently ignored.
**overridesobject= {}Keyword overrides forwarded into
EfficientDetConfig.Returns
EfficientDetForObjectDetectionDetector with the D0 configuration applied (or with overrides
merged on top of it).
Notes
See Tan et al., "EfficientDet: Scalable and Efficient Object Detection", CVPR 2020 (arXiv:1911.09070). D0 is the smallest of the eight compound-scaled variants and the recommended baseline for mobile / edge deployment.
Examples
>>> import lucid
>>> from lucid.models.vision.efficientdet import efficientdet_d0
>>> model = efficientdet_d0()
>>> x = lucid.randn(1, 3, 512, 512)
>>> out = model(x)
>>> out.pred_boxes.shape[-1]
4