efficientdet_d7(pretrained: bool = False, overrides: object = {})EfficientDet-D7 ().
The largest compound-scaled variant: EfficientNet-B6 backbone (shared with D6), BiFPN 384 channels with 8 repeats, head depth 5, and a 1536x1536 input. Approximately 51.9M parameters; COCO test-dev mAP of 53.7% (paper Table 2) — the headline accuracy result.
Model Size
Parameters
pretrainedbool= FalseReserved for future pretrained-weight loading. Currently ignored.
**overridesobject= {}Keyword overrides forwarded into
EfficientDetConfig.Returns
EfficientDetForObjectDetectionDetector with the D7 configuration applied.
Notes
See Tan et al., 2020 (arXiv:1911.09070), Table 2. The only architectural difference from D6 is the larger input resolution; backbone / BiFPN / head shapes are shared.
Examples
>>> import lucid
>>> from lucid.models.vision.efficientdet import efficientdet_d7
>>> model = efficientdet_d7()
>>> x = lucid.randn(1, 3, 1536, 1536)
>>> out = model(x)
>>> out.pred_boxes.shape[-1]
4