uaisoftwareinc/ai-3d-mapanything repository overview

⁠Map Anything - Docker

Docker implementation of Map Anything⁠.

⁠Overview

This Docker container runs the Map Anything gradio application on port 5000. Memory Efficient is set to True so inference can be performed on RTX 3090 and similar GPUs.

⁠Docker Build Targets

The Dockerfile uses multi-stage builds with two targets:

⁠Base Stage (base)

• Built on uaisoftwareinc/ai-cuda:latest
• Clones and installs the Map Anything repository
• Sets up Python virtual environment at /opt/venv
• Exposes port 5000
• Working directory: /data

⁠Development Target (dev)

docker build --target dev -t map-anything:dev .
docker run -it --gpus all -p 5000:5000 -v $(pwd):/data map-anything:dev

• Inherits from base
• Runs with sleep infinity to keep container alive
• Ideal for development and debugging
• Mount your local code to /data for live editing

⁠Production Target (prod)

docker build --target prod -t map-anything:prod .
docker run -d --gpus all -p 5000:5000 map-anything:prod

• Inherits from base
• Copies application code into the container
• Automatically runs the gradio application (scripts/gradio_app.py)
• Optimized for deployment

⁠Usage

⁠Development

Use the dev target when you need to modify code or debug:

# Build development image
docker build --target dev -t map-anything:dev .

# Run with volume mount for live code changes
docker run -it --gpus all -p 5000:5000 -v $(pwd):/data map-anything:dev bash

⁠Production

Use the prod target for deployment:

# Build production image
docker build --target prod -t map-anything:prod .

# Run production container
docker run -d --gpus all -p 5000:5000 map-anything:prod

The gradio interface will be available at http://localhost:5000.