SEAL is a PyTorch-based attribute learning package designed to facilitate the development and evaluation of attribute learning models. SEAL is designed to offer a flexible and modular framework for building attribute learning models. It leverages semantic information and uses state-of-the-art techniques to enhance the accuracy and interpretability of the learned attributes.
August 27, 2023: SEAL support distributed inference. We add text retrival image task.
August 24, 2023: One paper accepted in ACM MM 2023: Hierarchical Visual Attribute Learning in the Wild. The relevant code is now available. Please see project osarn.
August 18, 2023: Add HVAW dataset in seal/dataset/hvaw.py. Add new evaluation metric CV, CmAP and update the evaluation system.
Distributed Mode: We will soon update distributed training.
A Simple Tutorial: Allow user to quick add their own function based on SEAL.
| Model Name | Project | Status |
|---|---|---|
| Vision-language Guided Selective Loss | Vision-Language Assisted Attribute Learning | ✅ |
| Knowledge Enhanced Selective Loss | Attribute Learning with Knowledge Enhanced Partial Annotations | 🏗️ |
| Object-specific Attribute Relation Net | Hierarchical Visual Attribute Learning in the Wild | ✅ |
Here's how you can get started with SEAL:
- Clone the repo and install:
git clone https://github.com/PRIS-CV/seal.git
cd /path/to/seal
pip install -e .
- Import the models and start building attribute learning pipelines.
A brief architecture overview assists users in quickly grasping the structure of SEAL.
🏗️
Before running you should check the modular json settings in a project's directory, e.g., projects/gsl and see the running instruction in each project's README file:
CUDA_VISIBLE_DEVICES=0 python main.py --project projects/gsl --mode train
CUDA_VISIBLE_DEVICES=0 python main.py --project projects/gsl --mode test
If you use SEAL in your research or project, please consider citing the relevant papers.