MobileNetv2

MobileNetv2 leverages an inverted residual structure to build an efficient backbone.

We provide training and evaluation code of MobileNetv2, along with pretrained models and configuration files for the following tasks:

Image classification on the ImageNet dataset

Training

To train MobileNetv2 model on ImageNet using a single node with 4 A100 GPUs, run the following command:

export CFG_FILE="projects/mobilenet_v2/classification/mobilenetv2_1.0_in1k.yaml"
corenet-train --common.config-file $CFG_FILE --common.results-loc classification_results

We assume that the training and validation data is located in /mnt/imagenet/training and /mnt/imagenet/validation folders, respectively.

Evaluation

To evaluate the pre-trained MobileNetv2 model on the validation set of the ImageNet, run the following command:

export CFG_FILE="projects/mobilenet_v2/classification/mobilenetv2_1.0_in1k.yaml"
export DATASET_PATH="/mnt/vision_datasets/imagenet/validation/" # change to the ImageNet validation path
export MODEL_WEIGHTS=https://docs-assets.developer.apple.com/ml-research/models/cvnets-v2/classification/mobilenetv2-1.00.pt
CUDA_VISIBLE_DEVICES=0 corenet-eval --common.config-file $CFG_FILE --model.classification.pretrained $MODEL_WEIGHTS --common.override-kwargs dataset.root_val=$DATASET_PATH

This should give

top1=72.926 || top5=91.174

Semantic Segmentation using DeepLabv3 on the ADE20K dataset

Training

To train DeepLabv3-MobileNetv2 model on ADE20K using a single node with 4 A100 GPUs, run the following command:

export CFG_FILE="projects/mobilenet_v2/segmentation/deeplabv3_ade20k.yaml"
corenet-train --common.config-file $CFG_FILE --common.results-loc segmentation_results

We assume that the training and validation datasets are located in /mnt/vision_datasets/ADEChallengeData2016/ directory.

Evaluation

To evaluate the pre-trained DeepLabv3-MobileNetv2 model on the validation set of ADE20k, run the following command:

export CFG_FILE="projects/mobilenet_v2/segmentation/deeplabv3_ade20k.yaml"
export DATASET_PATH="/mnt/vision_datasets/ADEChallengeData2016/" # change to the ADE20k's path
export MODEL_WEIGHTS=https://docs-assets.developer.apple.com/ml-research/models/cvnets-v2/segmentation/ade20k/deeplabv3-mobilenetv2.pt
CUDA_VISIBLE_DEVICES=0 corenet-eval-seg --common.config-file $CFG_FILE --model.segmentation.pretrained $MODEL_WEIGHTS --common.override-kwargs dataset.root_val=$DATASET_PATH

This should give

mean IoU: 35.20

Pretrained Models

Classification (ImageNet-1k)

Model	Parameters	Top-1	Pretrained weights	Config file	Logs
MobileNetv2-0.25	1.5 M	53.57	Link	Link	Link
MobileNetv2-0.5	2.0 M	65.28	Link	Link	Link
MobileNetv2-0.75	2.6 M	70.42	Link	Link	Link
MobileNetv2-1.00	3.5 M	72.93	Link	Link	Link

Segmentation (ADE20k)

Note: The number of parameters reported does not include the auxiliary branches.

Model	Parameters	mIoU	Pretrained weights	Config file	Logs
DeepLabv3 MobileNetv2	8.0 M	35.20	Link	Link	Link

Citation

If you find our work useful, please cite following papers:

@article{Sandler2018MobileNetV2IR,
  title={MobileNetV2: Inverted Residuals and Linear Bottlenecks},
  author={Mark Sandler and Andrew G. Howard and Menglong Zhu and Andrey Zhmoginov and Liang-Chieh Chen},
  journal={2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2018},
  pages={4510-4520},
  url={https://api.semanticscholar.org/CorpusID:4555207}
}

@inproceedings{mehta2022cvnets, 
     author = {Mehta, Sachin and Abdolhosseini, Farzad and Rastegari, Mohammad}, 
     title = {CVNets: High Performance Library for Computer Vision}, 
     year = {2022}, 
     booktitle = {Proceedings of the 30th ACM International Conference on Multimedia}, 
     series = {MM '22} 
}