MobileNetv2 in PyTorch

An implementation of MobileNetv2 in PyTorch. MobileNetv2 is an efficient convolutional neural network architecture for mobile devices. For more information check the paper: Inverted Residuals and Linear Bottlenecks: Mobile Networks for Classification, Detection and Segmentation

Usage

Clone the repo:

git clone https://github.com/Randl/MobileNetV2-pytorch
pip install -r requirements.txt

Use the model defined in model.py to run ImageNet example:

python imagenet.py --dataroot "/path/to/imagenet/"

To run continue training from checkpoint

python imagenet.py --dataroot "/path/to/imagenet/" --resume "/path/to/checkpoint/folder"

Results

For x1.0 model I achieved 0.3% higher top-1 accuracy than claimed.

|Classification Checkpoint| MACs (M) | Parameters (M)| Top-1 Accuracy| Top-5 Accuracy| Claimed top-1| Claimed top-5| |-------------------------|------------|---------------|---------------|---------------|---------------|---------------| | [mobilenet_v2_1.0_224]|300 |3.47 | 72.1| 90.48| 71.8| 91.0|

You can test it with

python imagenet.py --dataroot "/path/to/imagenet/" --resume "results/mobilenet_v2_1.0_224/model_best.pth.tar" -e

• TODO: x0.35 model

• TODO: x1.4 model

• TODO: 96 input size model

• TODO: INT8 model (pytorch)