Pretrained weights for lucid.models.mobilenet_v4_hybrid_medium_cls.
Qin et al. ECCV 2024 Hybrid-Medium (11.1 M params, top-1 81.48%).
Attributes
IX_E550_R256_IN1KWeightEntryImageNet-1k checkpoint trained 550 epochs at 256x256 (top-1
81.478% / top-5 95.692% at 256, per timm's
results-imagenet.csv for this exact tag), sourced from
timm/mobilenetv4_hybrid_medium.ix_e550_r256_in1k. timm's
own default tag for this variant is pre-trained on ImageNet-12k;
this is its ImageNet-1k-only checkpoint at the paper's
resolution.DEFAULTWeightEntryAlias for
IX_E550_R256_IN1K.Notes
Reference: Qin et al., "MobileNetV4: Universal Models for the Mobile Ecosystem", ECCV 2024 (arXiv:2404.10518).
Examples
>>> from lucid.models.weights import MobileNetV4HybridMediumWeights
>>> list(MobileNetV4HybridMediumWeights.__members__)
['IX_E550_R256_IN1K', 'DEFAULT']
>>> MobileNetV4HybridMediumWeights.DEFAULT.meta["metrics"]["ImageNet-1k"]
{'acc@1': 81.478, 'acc@5': 95.692}