Docker

Docker is an open-source platform that enables developers to automate the deployment, scaling, and management of applications using containerization. In the context of artificial intelligence and machine learning, Docker acts as a standardized unit of software that packages code and all its dependencies—such as libraries, system tools, and settings—so the application runs quickly and reliably from one computing environment to another. This eliminates the common "it works on my machine" problem, ensuring that a neural network trained on a researcher's laptop behaves exactly the same way when deployed to a massive cloud server or an edge device.

Why Docker Matters for AI and Machine Learning

Modern machine learning operations (MLOps) rely heavily on reproducibility and portability. An AI project often involves a complex stack of software, including specific versions of Python, CUDA drivers for GPU acceleration, and deep learning frameworks like PyTorch or TensorFlow. Managing these manually across different teams and infrastructure is error-prone.

Docker simplifies this by creating lightweight, standalone containers. Unlike traditional virtual machines (VMs) that require a full operating system for each instance, containers share the host machine's OS kernel but run in isolated user spaces. This makes them significantly more resource-efficient and faster to start, which is critical when scaling model serving infrastructure or running distributed training jobs.

実際のアプリケーション

Docker is ubiquitous in the AI lifecycle, from initial experimentation to final deployment.

1. Consistent Training Environments: A data science team might use Docker images to share a unified development environment. For instance, a researcher working on object detection can pull a pre-built image containing all necessary drivers and libraries. This ensures that when they train a YOLO26 model, the results are reproducible by their colleagues, regardless of the underlying hardware differences.
2. Edge AI Deployment: In smart city surveillance, updated models need to be pushed to thousands of edge devices, such as traffic cameras or drones. Docker containers allow engineers to package a new model version and deploy it over the air. Because the container includes the inference runtime, the update process is seamless and does not interfere with the device's core operating system.

Docker vs. Kubernetes vs. Virtual Machines

It is helpful to distinguish Docker from related technologies to understand its specific role:

• Docker vs. Virtual Machines (VMs): VMs virtualize the hardware, meaning each VM runs a full OS (like Windows or Linux) on top of a hypervisor. This consumes significant memory and CPU. Docker virtualizes the operating system, making containers much smaller and faster than VMs.
• Docker vs. Kubernetes: These are complementary, not competing, technologies. Docker is the tool used to create and run individual containers. Kubernetes is a container orchestration platform that manages clusters of Docker containers, handling tasks like auto-scaling, load balancing, and self-healing in large-scale cloud computing environments.

例：コンテナ内での推論の実行

The following example demonstrates how a Python script might look inside a Docker container designed for computer vision. This script uses the ultralytics package to load a model and perform inference. The container environment ensures that the correct dependencies (like opencv-python そして torch) are already present.

from ultralytics import YOLO

# Load the YOLO26 model (weights are typically included in the Docker image)
model = YOLO("yolo26n.pt")

# Perform inference on an image source
# In a containerized microservice, this might process incoming API requests
results = model.predict("https://ultralytics.com/images/bus.jpg")

# Log the detection results
print(f"Detected {len(results[0].boxes)} objects in the image.")

Integrating Docker into Your Workflow

To get started with containerization, developers typically define a Dockerfile, which is a text document containing all the commands to assemble an image. Once built, these images can be stored in registries like Docker Hub or the NVIDIA NGC Catalog, which offers GPU-optimized containers.

For those looking to streamline the training and deployment process without manually managing Dockerfiles, the Ultralytics Platform offers integrated tools that handle the complexity of cloud environments. This allows users to focus on improving model accuracy rather than configuring infrastructure. Furthermore, you can explore our Docker Quickstart Guide to learn how to run Ultralytics models in containers immediately.