Adaptive and Multi-scale Affinity Alignment for Hierarchical Contrastive Learning

codebase for experiments related to Adaptive and Multi-scale Affinity Alignment.

Project Overview

This repository contains the current PyTorch implementation for AMA-style hierarchical contrastive learning experiments.

Installation

Create an environment with Python 3.8+ and install the required packages:

pip install -r requirements.txt

The current requirements are:

• torch>=1.8.0
• torchvision>=0.9.0
• tensorboard>=2.0.0

Configuration

major hyperparameters are defined in config.py, including:

• Backbone architecture
• Feature dimension
• Queue size
• Multi-scale clustering settings
• Optimizer settings
• Batch size and number of epochs
• Dataset selection