FD-MobileNet: Improved MobileNet with A Fast Downsampling Strategy

Introduction

This repository contains the code for FD-MobileNet (Fast-Downsampling MobileNet), an efficient and accurate network for very limited computational budgets (e.g., 10-140 MFLOPs).

Our paper FD-MobileNet: Improved MobileNet with A Fast Downsampling Strategy is under review as a conference paper at ICIP 2018.

For details, please refer to our arXiv version.

Requirements

1. pytorch >= 0.2.0, torchvision >= 0.2.0
2. graphviz >= 0.8.0

Usage

To train a model:

python -u main.py \
       --data /path/to/data/config \
       --model /path/to/model/config \
       --optim /path/to/optim/config \
       --sched /path/to/sched/config \
       --label model_label \
       [--print-freq N] \
       [--resume] \
       [--evaluate]

where model_label is the name of the checkpoint to be saved or resumed. For example:

python -u main.py \
       --data config/imagenet/data-config/imagenet-aggressive.json \
       --model config/imagenet/model-config/fd-mobilenet/1x-FDMobileNet-224.json \
       --optim config/imagenet/optim-config/SGD-120-nesterov.json \
       --sched config/imagenet/sched-config/StepLR-30-0.1.json \
       --label 1x-FDMobileNet-224

For simplicity, we train models and save checkpoints in multi-GPU models (using torch.nn.DataParallel), which means the keys in the state_dict saved have the prefix module.. To convert a multi-GPU model to single-GPU model, run convert_model.py:

python -u convert_model.py \
       --data /path/to/data/config \
       --model /path/to/model/config \
       --label model_label \
       --input /path/to/checkpoint/file \
       --output /path/to/output/file

Our pre-trained models are single-GPU models (without prefix). To evaluate single-GPU models, run evaluate.py:

python -u evaluate.py \
       --data /path/to/data/config \
       --model /path/to/model/config \
       --checkpoint /path/to/checkpoint/file \
       [--print-freq N]

main.py is modified from the pytorch example.

Results on ImageNet 2012

Training configuration:

1. 120 epochs;
2. Batch size: 256;
3. Momentum: 0.9;
4. Weight decay: 4e-5;
5. Learning rate starts from 0.1, divided by 10 every 30 epochs;
6. 4 GPUs.

Model	MFLOPs	Top-1 Acc. (%)
FD-MobileNet 1$\times$	144	65.3
FD-MobileNet 0.5$\times$	40	56.3
FD-MobileNet 0.25$\times$	12	45.1