Giter Site home page Giter Site logo

luoyadan / sf-pgl Goto Github PK

View Code? Open in Web Editor NEW
21.0 2.0 4.0 85.14 MB

This work is the official Pytorch implementation of our papers: Source-Free Progressive Graph Learning for Open-Set Domain Adaptation (TPAMI))

License: MIT License

Python 97.21% Shell 2.79%
source-free domain-adaptation open-set-domain-adaptation progressive-learning

sf-pgl's Introduction

SF-PGL

This work is the official Pytorch implementation of our papers:

Source-Free Progressive Graph Learning for Open-Set Domain Adaptation
Yadan Luo, Zijian Wang, Zhuoxiao Chen, Zi Huang, Mahsa Baktashmotlagh
Transcations on Pattern Analysis and Machine Intelligence (TPAMI)

Progressive Graph Learning for Open-Set Domain Adaptation
Yadan Luo^, Zijian Wang^, Zi Huang, Mahsa Baktashmotlagh
International Conference on Machine Learning (ICML) 2020
[Paper] [Code]


Framework

To further handle a more realistic yet challenging source-free setting, a novel SF-PGL framework was proposed, which leverages a balanced pseudo-labeling regime to enable uncertainty-aware progressive learning without relying on any distribution matching or adversarial learning methods. As an extension of PGL, we have significantly extended the idea of open-set domain adaptation from the unsupervised learning case to the source-free and semi-supervised settings, from image classification to action recognition, where the complex data interaction and more significant domain gap are addressed. We further discussed a hitherto untouched aspect of OSDA model - the model calibration issue. Experimental results evidenced that the SF-PGL can alleviate the class imbalance introduced by pseudo-labeled sets so that the overconfidence and under-confidence of the OSDA model can be avoided.


Contents

Requirements

  • Python 3.6
  • Pytorch 1.3

Dataset Preparation

  • Office-home
  • VisDA-17
  • Syn2Real-O (VisDA-18)

Training

The general command for training is,

python3 train.py

Change arguments for different experiments:

  • dataset: "home" / "visda" / "visda18"
  • batch_size: mini_batch size
  • beta: The ratio of known target sample and Unk target sample in the pseudo label set
  • EF : Enlarging Factor α
  • num_layers: GNN's depth
  • adv_coeff: adversarial loss coefficient γ
  • node_loss: node classification loss μ For the detailed hyper-parameters setting for each dataset, please refer to Section 5.2 and Appendix 3.

Remember to change dataset_root to suit your own case

The training loss and validation accuracy will be automatically saved in './logs/', which can be visualized with tensorboard. The model weights will be saved in './checkpoints'

sf-pgl's People

Contributors

anom1458 avatar busername avatar luoyadan avatar

Stargazers

 avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar

Watchers

 avatar  avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.