Towards Effective Deep Transfer via Attentive Feature Alignment

We provide PyTorch implementation for "Towards Effective Deep Transfer via Attentive Feature Alignment" (NN2021).

Dependencies

• Python 3.6
• Pytorch 1.1.0
• Dependencies in requirements.txt

Installation

1. Clone this repo:

    git clone https://github.com/xiezheng-cs/AFA.git
    cd AFA
   

2. Install pytorch and other dependencies.

    pip install -r requirements.txt
   

Training

1. First Stage:

    ResNet101: python main.py hocon_config/First_stage_AST/resnet101_stanford_dogs_120.hocon
   
    MobileNet_V2: python main.py hocon_config/First_stage_AST/mobilenet_v2_stanford_dogs_120.hocon
   

2. Second Stage:

    ResNet101: python main.py hocon_config/Second_stage_ACT/resnet101_Stanford_dogs_120.hocon
   
    MobileNet_V2: python main.py hocon_config/Second_stage_ACT/mobilenet_v2_Stanford_dogs_120.hocon
   

Results

Target Data Set	Model	DELTA Top1 Acc(%)	AFA(Ours) Top1 Acc(%)
Stanford Dogs 120	MobileNetV2	81.3±0.1	82.1±0.1
Stanford Dogs 120	ResNet-101	88.7±0.1	90.1±0.0

Pre-trained Model

Model	Link	Top1 Acc (%)
ResNet101	https://github.com/xiezheng-cs/AFA/releases/tag/models	90.22
MobileNetV2	https://github.com/xiezheng-cs/AFA/releases/tag/models	82.17

Eval

    ResNet101: python main.py hocon_config/val/resnet101_Stanford_dogs_120.hocon

    MobileNet_V2: python main.py hocon_config/val/mobilenet_v2_Stanford_dogs_120.hocon

Citation

If this work is useful for your research, please cite our paper:

@InProceedings{xie2021afa,
title = {Towards Effective Deep Transfer via Attentive Feature Alignment},
author = {Zheng Xie, Zhiquan Wen, Yaowei Wang, Qingyao Wu, and Mingkui Tan},
journal = {Neural Networks (NN)},
volume = {138},
pages = {98-109},
year = {2021}
}