🌍 OpenHAIV

A Framework for Open-World Learning

[Manual] [Arxiv] [Dataset] [Forum]

Star to bookmark, fork & cite if you use, and welcome to contribute!

👋 Welcome to OpenHAIV

OpenHAIV is a comprehensive framework for open-world object recognition, supporting out-of-distribution detection, novel class discovery, and incremental learning. It is designed to enable robust and flexible object recognition in unconstrained environments.

The framework adopts a modular design overall, which is reflected in two key aspects.

• 🛠️ Functionally: The framework independently incorporates dedicated modules for supervised training, out-of-distribution detection, novel class discovery, and incremental learning.
• ⚙️ Procedurally：The framework divides its operational workflow into distinct stages, including data processing, model construction, training and evaluation, and visualization.

⭐ Key Features

🧩 Modular Design

• Easy experimentation with different algorithms
• Seamless integration of new techniques
• Flexible deployment configurations

🚀 Scalable Architecture

• Large-scale dataset processing
• Distributed training capabilities
• Efficient memory management
• Parallel processing optimization

🔌 Extensible Framework

• Plugin-based algorithm integration
• Custom loss function support
• Flexible evaluation metrics
• Comprehensive logging and monitoring

📑 Table of Contents

• 👋 Welcome to OpenHAIV
• 📑 Table of Contents
• 🎉 News
• 📋 TODO List
• 🚀 Getting Started
  • 📦 Installation
  • 🔧 Environment Setup
  • 🏃‍♂️ Running Examples
• 🤖 Supported Models
• 📚 Supported Methods
  • 🌱 Class-incremental learning
  • 🚨 Out-of-Distribution Detection
  • 🔍 Novel Class Discovery
  • 🧬 Data Augmentation
• 🛠️ Contributing Guidelines
  • 🧹 Code Style
  • 📜 Code of Conduct
  • 🐛 Reporting Issues
  • 💡 Submitting Pull Requests
• 🤝Contributors
• 📖Citation
• 🙏Acknowledgement
• ✉️Contact

🎉 News

[08/2025] 🌟 Our paper "OpenHAIV: A Framework Towards Practical Open-World Learning" is available at [Arxiv]. [08/2025] 🌟 The code repository is is now publicly available. [09/2024] 🌟 The code repository is created.

📋 TODO List

• Implement core framework structure
• Support basic OOD detection methods
• Support basic class-incremental learning methods
• Complete Novel Class Discovery implementation
• Add comprehensive benchmarks and evaluation metrics
• Add Data Augmentation techniques
• Create detailed documentation and tutorials
• Add visualization tools for analysis
• Support Windows & macOS installation guide
• Create Jupyter notebook examples
• Set up continuous integration for testing
• Add pre-trained model zoo
• Implement unified evaluation protocols

🚀 Getting Started

📦 Installation

It is recommended to use anaconda3 to manage and maintain the python library environment.

1. Download the .sh file from the anaconda3 website
2. Install anaconda3 with .sh file

bash Anaconda3-2023.03-Linux-x86_64.sh

🔧 Environment Setup

Create and activate a virtual environment:

conda create -n openhaiv python=3.10 -y
conda activate openhaiv
pip install -r requirements.txt
python setup.py install

Required packages:

• pytorch>=1.12.0 torchvision>=0.13.0 (recommend official torch command)
• numpy>=1.26.4
• scipy>=1.14.0
• scikit-learn>=1.5.1

🏃‍♂️ Running Examples

Out-of-Distribution Detection

python ncdia/train.py --cfg configs/pipeline/ood_detection/msp/det_oes_rn50_msp_train.yaml --opts device='cuda:0'

Class-incremental Learning

bash ./scripts/inc_BM200_lwf.sh

Novel Class Discovery

# Set required parameters
# - model weight in weight_path
# - id_txt_file and ood_txt_file
# - output_dir

python ncd.py

🤖 Supported Models

ResNet、ViT、CLIP

📚 Supported Methods

🌱 Class-incremental learning

CNN-based methods

Method	Paper	Venue
Joint	update models using all the data from all classes
Finetune	baseline method which simply updates model using current data
LwF	Learning without Forgetting	ECCV 2016
EWC	Overcoming catastrophic forgetting in neural networks	PNAS 2017
iCaRL	Incremental Classifier and Representation Learning	CVPR 2017
BiC	Large Scale Incremental Learning	CVPR 2019
WA	Maintaining Discrimination and Fairness in Class Incremental Learning	CVPR 2020

|GEM|Gradient Episodic Memory for Continual Learning|NIPS 2017| |SSRE|Self-Sustaining Representation Expansion for Non-Exemplar Class-Incremental Learning|CVPR 2022|

ViT-based methods

Method	Paper	Venue
Joint	update models using all the data from all classes
Finetune	baseline method which simply updates model using current data

CLIP-based methods

Method	Paper	Venue
Joint	update models using all the data from all classes
Finetune	baseline method which simply updates model using current data

Few-shot class-incremental learning

Method	Paper	Venue
Alice	Few-Shot Class-Incremental Learning from an Open-Set Perspective	ECCV 2022
FACT	Forward Compatible Few-Shot Class-Incremental Learning	CVPR 2022
SAVC	Semantic-Aware Virtual Contrastive Constraint for Few-Shot Class-Incremental Learning	CVPR 2023

🚨 Out-of-Distribution Detection

CNN-based Methods

Method	Paper	Venue
MSP	A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks	ICLR 2017
ODIN	Enhancing The Reliability of Out-of-distribution Image Detection in Neural Networks	ICLR 2018
MDS	A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks	NeurIPS 2018
MLS	Scaling Out-of-Distribution Detection for Real-World Settings	ICML 2022
ViM	Out-Of-Distribution with Virtual-logit Matching	CVPR 2022
FDBD	Fast Decision Boundary based Out-of-Distribution Detector	ICML 2024
VOS	Learning What You Don't Know by Virtual Outlier Synthesis	ICLR 2022
LogitNorm	Mitigating Neural Network Overconfidence with Logit Normalization	ICML 2022
DML	Decoupling MaxLogit for Out-of-Distribution Detection	CVPR 2023

CLIP-based Methods

Method	Paper	Venue
MCM	Delving into Out-of-Distribution Detection with Vision-Language Representations	NeurIPS 2022
GLMCM	Global and Local Maximum Concept Matching for Zero-Shot Out-of-Distribution Detection	IJCV 2025
CoOp	Learning to Prompt for Vision-Language Models	IJCV 2022
LoCoOp	Few-Shot Out-of-Distribution Detection via Prompt Learning	NeurIPS 2023
SCT	Self-Calibrated Tuning of Vision-Language Models for Out-of-Distribution Detection	NeurIPS 2024
DPM	Vision-Language Dual-Pattern Matching for Out-of-Distribution Detection	ECCV 2024

🔍 Novel Class Discovery

TBD

🧬 Data Augmentation

TBD

🛠️ Contributing Guidelines

We welcome contributions to OpenHAIV🤗 If you're interested in improving the project, please follow these guidelines:

🧹 Code Style

This project uses pre-commit to automatically enforce code style and quality before each commit. Please install pre-commit and run:

pip install pre-commit
pre-commit install

The main checks include:

• flake8: PEP8 code style checking
• yapf: automatic Python code formatting
• codespell: spell checking
• docformatter: automatic docstring formatting
• trailing-whitespace, end-of-file-fixer, mixed-line-ending and other basic formatting fixes

See the .pre-commit-config.yaml file for detailed configuration. All these checks and fixes will be run automatically before every commit.

Note: After running pre-commit install, the configured hooks will automatically run on every git commit. It is recommended to run pre-commit run --all-files once to fix and harmonize existing files in the repository before making commits. To skip hooks for a specific commit, use git commit --no-verify (not recommended).

📜 Code of Conduct

Please note that all contributors are expected to follow our Code of Conduct to foster a welcoming and inclusive community.

🐛 Reporting Issues

1. Check existing issues first to avoid duplicates
2. Use the issue template when available
3. Be specific about the problem:
   • Include steps to reproduce
   • Provide environment details (OS, Python version, dependencies)
   • Add screenshots if applicable
   • Describe expected vs. actual behavior

💡 Submitting Pull Requests

1. Create an issue first to discuss major changes
2. Fork the repository and create a branch from main
3. Follow the coding style used throughout the project:
   • Adhere to PEP 8 guidelines
   • Use meaningful variable/function names
   • Add docstrings for new functions/classes
4. Write tests for new features
5. Ensure all tests pass before submitting
6. Update documentation reflecting your changes
7. Make atomic commits with clear messages

🤝Contributors

Xiang Xiang, Qinhao Zhou, Jing Ma, Zhuo Xu, Jiaxin Dai, Yifan Liang, Hanlin Li, Yao Deng, Zhipeng Chen, Zihan Zhang, Yuwen Tan.

📖Citation

If you find our repository useful for your research, please consider citing these papers:

@article{xiang2025openhaiv,
  title={OpenHAIV: A Framework Towards Practical Open-World Learning},
  author={Xiang, Xiang and Zhou, Qinhao and Xu, Zhuo and Ma, Jing and Dai, Jiaxin and Liang, Yifan and Li, Hanlin},
  journal={arXiv preprint arXiv:2508.07270},
  year={2025},
  url={https://arxiv.org/abs/2508.07270},
  doi={10.48550/arXiv.2508.07270}
}

🙏Acknowledgement

• OpenOOD, an extensible codebase for out-of-distribution detection with Vision Models only.
• OpenOOD-VLM, an extensible codebase for out-of-distribution detection with both Vision Models and Vision-Language Models.
• PyCIL, an extensible codebase for incremental learning.

✉️Contact

If there are any questions, please feel free to propose new features by opening an issue or contact with the team leader: Xiang Xiang (xex@hust.edu.cn). Enjoy the code.