Locke0 / episodic-memory-object-clustering

Episodic Memory Object Clustering Project

Overview

This repository contains code and resources for a project aimed at clustering distinct entities in egocentric videos ("episodic memory"). The primary goal is to develop a system that can analyze video content and identify separate entities or objects, allowing for meaningful clustering and classification.

For details on episodic memory, refer to this page.

Features

• Entity Clustering: The project leverages advanced computer vision and machine learning techniques to cluster entities within video frames.

• AWS Integration: AWS credentials are required for accessing cloud-based resources. Ensure to set up your AWS credentials in the .env file.

• Ego4D Annotations: The project utilizes Ego4D Annotations to enrich video data. The download_data.sh script facilitates the download of Ego4D Annotations and relevant benchmark repositories.

Setup

Prerequisites

• Miniconda: Install Miniconda to manage the project environment. Refer to the scripts/setup_environment.sh script for automated setup.

• AWS CLI: Set up the AWS CLI by running the scripts/setup_environment.sh script. AWS credentials are stored in the .env file.

• Ego4D CLI: Install the Ego4D CLI using pip install ego4d to access annotations and enhance video data.

Environment Setup

./scripts/setup_environment.sh

Download Data and Annotations

./scripts/download_data.sh

Usage

Environment Activation:

conda activate ego4d_vq2d

License

This project is licensed under the MIT License.

Acknowledgments

Ego4D: Special thanks to the Ego4D team for providing the dataset