EfficientNetV2: Smaller Models and Faster Training
Paper • 2104.00298 • Published • 1
LibreEfficientNetV2b3-cls
EfficientNetV2-base-b3 image classifier (1000-class ImageNet-1k), repackaged for LibreYOLO. Eval resolution 300px; timm-reported top-1 accuracy ~82.1%.
Usage
from libreyolo import LibreYOLO
model = LibreYOLO("LibreEfficientNetV2b3-cls.pt")
result = model.predict("image.jpg")[0]
print(result.probs.top1, result.probs.top1conf)
print(result.probs.top5)
Source
Derived from the timm checkpoint
tf_efficientnetv2_b3.in1k in
huggingface/pytorch-image-models.
Copyright (c) 2019 Ross Wightman. Licensed under the Apache License 2.0.
Original architecture: EfficientNetV2 by Google (google/automl), "EfficientNetV2: Smaller Models and Faster Training" (arXiv:2104.00298), Apache License 2.0. Only the ImageNet-1k checkpoint is published here — the ImageNet-21k / JFT variants carry extra-data terms and are intentionally excluded.
Modifications
State-dict key remapping only. Learned parameters are unchanged; inference is
bit-identical to timm (max_abs_diff == 0). See
weights/convert_efficientnetv2_weights.py in the
LibreYOLO source repository.
License
Apache License 2.0. See the LICENSE and NOTICE
files in this repository.