Attention-based Sampler in TASN (Trilinear Attention Sampling Network)

It is an implemetation of attention-based sampler in TASN.

It's based on MobulaOP, and you don't need to re-build MXNet.

In addition, the implementation of attention-based sampler is available for MXNet and PyTorch.

Usage

1. Install MobulaOP

# Clone the project
git clone https://github.com/wkcn/MobulaOP

# Enter the directory
cd MobulaOP

# Install MobulaOP
pip install -v -e .

2. Clone TASN project

git clone https://github.com/researchmm/tasn
cd tasn/tasn-mxnet/example/tasn

3. Clone this project

git clone https://github.com/wkcn/AttentionSampler

The directory shows as follow:

├─AttentionSampler
│   ├── attention_sampler
│   ├── imgs
│   └── test.py
├── common
├── data
├── init.sh
├── install.sh
├── model
├── model.py
├── readme
├── train.py
└── train.sh

4. Copy the following code on the head of model.py of TASN

import mxnet as mx
import mobula
from AttentionSampler.attention_sampler import attsampler_mx
mobula.op.load('./AttentionSampler/attention_sampler')

You can train TASN model now. Enjoy it!

If this project is helpful, Hope to follow me and star the MobulaOP project.

Thank you!

Training Log

Log (default setting)

Reference Paper

@inproceedings{zheng2019looking,
  title={Looking for the Devil in the Details: Learning Trilinear Attention Sampling Network for Fine-grained Image Recognition},
  author={Zheng, Heliang and Fu, Jianlong and Zha, Zheng-Jun and Luo, Jiebo},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={5012--5021},
  year={2019}
}