BattleshipGraphicsProjects

An umbrella repository for the experiments and tooling that turn historical warship illustrations into 3D-ready assets. It groups together several independent sub-projects (image extraction, image-to-3D pipelines, a Unity MCP integration, and reference scans) under one tree.

The two big pillars right now are:

1. Florence-2 Warship Extractor (this repo's src/) — extract warship illustrations from historical PDFs.
2. ProjectBroadside — the downstream pipelines that take those images and produce 3D meshes / Unity content.

Repository layout

Path	What it is
src/warship_extractor/	Python package: Florence-2-based extractor (cli.py, pipeline/, detection/, processing/, core/, config/, utils/).
tests/	Pytest suite (unit/, integration/) for the warship extractor.
pyproject.toml / poetry.lock	Poetry config for the extractor. Python >=3.9,<3.12.
FLORENCE_2_WARSHIP_EXTRACTION_PLAN.md	Architectural plan for the extractor.
Docs/	Cross-project documentation (2D pipeline, strategy notes, ComfyUI workflow prompts, archive).
Scans/	Source PDF / scanned material used as input.
BAttleships/	Reference warship images.
ProjectBroadside/	The 3D-asset side of the project (see below).

ProjectBroadside subprojects

Path	What it is
ProjectBroadside/BattleshipMaker/	First-iteration 2D→3D pipeline (Python; uses Gemini for view-detection / vectorization).
ProjectBroadside/BattleshipMaker2/	Reworked staged pipeline: image generation → dataset prep → 3DGS training → splat refinement → mesh conversion → final output.
ProjectBroadside/HullCutter/	Placeholder / stub for hull-extraction tooling.
ProjectBroadside/MCPUnityRockstar/	MCP Unity Editor integration (Node.js MCP server + Unity Editor package).
ProjectBroadside/PDFScans/	Scanned PDF inputs used by the 3D pipelines.
ProjectBroadside/ProjectBroadside.Scripts/	Unity C# scripts (gameplay, ECS, authoring, components, core, data).

Florence-2 Warship Extractor

A pipeline for pulling warship illustrations out of Jane's Fighting Ships and similar naval archives, using Microsoft's Florence-2 vision-language model.

Highlights

• Florence-2-Large for object detection
• Multi-prompt strategy to catch different illustration styles
• High-resolution PDF rasterization (300+ DPI by default)
• Non-Maximum Suppression to drop duplicate detections
• CUDA-aware with CPU fallback; dynamic batch sizing
• CLI (warship-extract) for extract / batch / report / info

Install

Requires Python 3.9–3.11 and Poetry. A CUDA-compatible GPU is recommended.

poetry install
poetry shell

CLI

# Extract from a single PDF
warship-extract extract input.pdf --output-dir results/

# Batch process a folder of PDFs
warship-extract batch pdf_folder/ --output-dir results/

# Generate an analysis report
warship-extract report results/ --format html

# System / environment info
warship-extract info

Python API

from warship_extractor.pipeline import ExtractionPipeline
from warship_extractor.config import Settings

settings = Settings()
pipeline = ExtractionPipeline(settings)
results = pipeline.extract_from_pdf("janes_1900.pdf")

for detection in results["detections"]:
    print(f"Found {detection['label']} with confidence {detection['confidence']:.2f}")

Configuration

The extractor reads settings from environment variables or Settings:

FLORENCE_MODEL_NAME="microsoft/Florence-2-large"
FLORENCE_DEVICE="cuda"
PDF_DPI=300
DETECTION_CONFIDENCE_THRESHOLD=0.7
OUTPUT_DIR="./extracted_warships"
SAVE_ANNOTATED_IMAGES=true

Tests

poetry run pytest
poetry run pytest --cov=src/warship_extractor
poetry run pytest tests/unit/test_detector.py

pyproject.toml already wires up pytest --cov against src/warship_extractor by default.

ProjectBroadside

ProjectBroadside/ is a heterogeneous workspace, not a single Python package:

• BattleshipMaker/ — earlier Python experiment that takes 2D ship views (side / top), splits multi-view images, and tries to produce a base 3D mesh. See BattleshipMaker/Prompt.md and framework_proposal.md.
• BattleshipMaker2/ — current staged pipeline using image generation, 3D Gaussian Splatting training, splat refinement, and mesh conversion. See BattleshipMaker2/README.md and config.yaml.
• MCPUnityRockstar/ — a Model Context Protocol server that exposes the Unity Editor as tools to MCP-aware clients (Node-based server in Server~/ plus a Unity Editor/ package).
• ProjectBroadside.Scripts/ — the Unity-side C# code (ECS, gameplay, authoring, components).

Each subproject has its own README / setup notes; this top-level README only orients you.

Status

Active, multi-project, exploratory. Sub-projects move at different speeds — treat the per-subproject documentation as the source of truth for whichever pipeline you're working in.