1 Star 0 Fork 0

hzw_ai / DSN

Create your Gitee Account
Explore and code with more than 6 million developers,Free private repositories !:)
Sign up
Clone or Download
Cancel
Notice: Creating folder will generate an empty file .keep, because not support in Git
Loading...
README.md

Domain-Smoothing Network for Zero-Shot Sketch-Based Image Retrieval

This project is our implementation of Domain-Smoothing Network (DSN) for Zero-Shot Sketch-Based Image Retrieval. The details in methods and experiments could be found in the paper

framework

If you find this project helpful, please consider to cite our paper:

@misc{wang2021domainsmoothing,
      title={Domain-Smoothing Network for Zero-Shot Sketch-Based Image Retrieval}, 
      author={Zhipeng Wang and Hao Wang and Jiexi Yan and Aming Wu and Cheng Deng},
      year={2021},
      eprint={2106.11841},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Dataset

The datasets we used are provided by SAKE. You can download the resized Sketchy Ext and TU-Berlin Ext dataset and train/test split files from here. Then unzip the datasets files to the same directory ./dataset of this project.

Training

CSE-ResNet50 model with 64-d features in default setting

# train with Sketchy Ext dataset
python train_cse_resnet_sketchy_ext.py

# train with TU-Berlin Ext dataset
python train_cse_resnet_tuberlin_ext.py

Testing

CSE-ResNet50 model with 64-d features in default setting

# test with Sketchy Ext dataset
python test_cse_resnet_sketchy_zeroshot.py

# test with TU-Berlin Ext dataset
python test_cse_resnet_tuberlin_zeroshot.py

Pre-trained Models

Our trained models for Skethy Ext and TU-Berlin Ext with 64-d features in default setting can be downloaded from here, passwd: DSN2. Please modify the file name of pre-trained models to model_best.pth.tar then put it to the corresponding directory in ./cse_resnet50/checkpoint/.

For example, the path of pre-trained model for Sketchy Ext in default experimental setting should be:

./cse_resnet50/checkpoint/sketchy_kd(1.0)_kdneg(0.3)_sake(1.0)_dim(64)_contrastive(128-0.1)_T(0.07)_memory(10-1.0)/model_best.pth.tar

Repository Comments ( 0 )

Sign in for post a comment

About

Pytorch Implementation of DSN (IJCAI 2021) expand collapse
Python
Apache-2.0
Cancel

Releases

No release

Contributors

All

Activities

Load More
can not load any more
1
https://gitee.com/hzw-ai/DSN.git
git@gitee.com:hzw-ai/DSN.git
hzw-ai
DSN
DSN
main

Search

141041 ab9339c7 1850385 141043 25c028d5 1850385