PyTorch Implementation of MobileNet V4

A PyTorch reproduction of the MobileNetV4 architecture has been published, based on the paper "MobileNetV4 - Universal Models for the Mobile Ecosystem" by Danfeng Qin, Chas Leichner, Manolis Delakis, Marco Fornoni, Shixin Luo, Fan Yang, Weijun Wang, Colby Banbury, Chengxi Ye, Berkin Akin, Vaibhav Aggarwal, Tenghui Zhu, Daniele Moro and Andrew Howard (arXiv:2404.10518). The release lists three model variants: MobileNetV4-S with 4.30M parameters and 0.306G FLOPs at 224 pixels, MobileNetV4-M with 9.72M parameters and 1.080G FLOPs at 256 pixels, and MobileNetV4-L with 32.59M parameters and 6.376G FLOPs at 384 pixels. The paper's official TensorFlow implementation remains available in the TensorFlow models repository.

Reproduction of MobileNet V4 architecture as described in MobileNetV4 - Universal Models for the Mobile Ecosystem https://arxiv.org/abs/2404.10518 by Danfeng Qin, Chas Leichner, Manolis Delakis, Marco Fornoni, Shixin Luo, Fan Yang, Weijun Wang, Colby Banbury, Chengxi Ye, Berkin Akin, Vaibhav Aggarwal, Tenghui Zhu, Daniele Moro, Andrew Howard with the PyTorch https://github.com/d-li14/mobilenetv4.pytorch/blob/main/pytorch.org framework. | Architecture | Parameters | FLOPs @ pix | Top-1 Acc. % | |---|---|---|---| | MobileNetV4-S | 4.30M | 0.306G @ 224 | | | MobileNetV4-M | 9.72M | 1.080G @ 256 | | | MobileNetV4-L | 32.59M | 6.376G @ 384 | | @misc{qin2024mobilenetv4, title={MobileNetV4 -- Universal Models for the Mobile Ecosystem}, author={Danfeng Qin and Chas Leichner and Manolis Delakis and Marco Fornoni and Shixin Luo and Fan Yang and Weijun Wang and Colby Banbury and Chengxi Ye and Berkin Akin and Vaibhav Aggarwal and Tenghui Zhu and Daniele Moro and Andrew Howard}, year={2024}, eprint={2404.10518}, archivePrefix={arXiv}, primaryClass={cs.CV} } The official TensorFlow implementation https://github.com/tensorflow/models/blob/master/official/vision/modeling/backbones/mobilenet.py .