mobilenet_v3_large(pretrained: bool = False, overrides: object = {})MobileNet-v3-Large feature-extracting backbone.
Builds a MobileNetV3 with the NAS-designed Large
topology from Howard et al., 2019 (Table 1): 15 inverted-residual
bottleneck blocks with selective SE attention and hard-swish in
the deeper half, followed by a 1×1 expansion to 960 channels.
Approximately 5.4M parameters and 75.2% ImageNet-1k top-1
accuracy (Table 3) — the higher-accuracy variant of the family.
Model Size
Parameters
pretrainedbool= FalseReserved for future pretrained-weight loading. Currently
ignored — the returned model is randomly initialised.
**overridesobject= {}Keyword overrides forwarded into
MobileNetV3Config
(e.g. width_mult=0.75 for a slimmer variant).Returns
MobileNetV3Backbone with the MobileNet-v3-Large configuration applied
(or with overrides merged on top of it).
Notes
See Howard et al., "Searching for MobileNetV3", ICCV 2019 (arXiv:1905.02244), Table 1 (V3-Large spec) and Table 3 (ImageNet accuracy).
Examples
>>> import lucid
>>> from lucid.models.vision.mobilenet_v3 import mobilenet_v3_large
>>> model = mobilenet_v3_large()
>>> x = lucid.randn(1, 3, 224, 224)
>>> out = model(x)
>>> out.last_hidden_state.shape
(1, 960, 1, 1)