Add Reproducible Environment Setup with Pinned Dependencies

[aragorn1025]

Overview

This PR introduces a Docker-based environment for DINO to simplify repository setup and provide a reproducible development environment by pinning compatible dependency versions.

Background

Some Python dependencies used in this repository are no longer backward compatible with their latest releases. As a result, setting up the project with newer package versions may lead to compatibility issues or unexpected behavior.

Issues

This PR addresses the following issues:

• TypeError: FormatCode() got an unexpected keyword argument 'verify' #232
• Incompatibility of the newest NumPy version during testing #253
• Error while training #257

Changes

• Add a Dockerfile that pins selected Python package versions for compatibility.
• Add a docker-compose.yml and a Makefile to simplify environment setup.
• Add docker-compose.override.yml to .gitignore to allow users to customize local configurations, such as dataset, pretrained weights, and checkpoint mounts, without committing them to the repository.
• Update .gitignore to exclude generated build files.

Testing

This setup has been tested on an Ubuntu Linux environment with NVIDIA GPU support.

The host system requires:

• NVIDIA GPU driver
• NVIDIA Container Toolkit

Other environments, such as CPU-only or non-Linux systems, have not been verified.

[aragorn1025]

This PR reopens #284 with an updated source branch.