Custom Parametric Shoe Trees

Foot scan to printable, custom-fit shoe tree.

By Alex, James, and Naijun for FABRIC-581.

Demo

If the embedded video does not render, open the demo video directly. The final presentation deck is also included at docs/presentation/Custom-Parametric-Shoe-Trees-Final-Presentation.pdf.

Problem

Traditional shoe trees are sold in "universal" sizes and expand at fixed regions such as the toe box or length. That mismatch can push a shoe away from the owner's actual foot shape, deforming the upper and accelerating structural wear.

Commercial shoe trees also cost about $20-$30. A scan-driven, 3D-printed version can be produced for a few dollars, with a shape that reflects the user's own foot anatomy. Athletic shoes are a strong use case because they deform after training and often need a controlled break-in shape.

Design Motivation

The project uses a parametric pipeline rather than a fixed-size catalog. Each shoe tree is driven by measurements from a foot scan, so length, ball width, toe height, arch position, heel center, and upper-surface allowance can change per person. This leaves room for future design features such as breathable lattice bodies, split print layouts, and customized snap-fit connections.

Pipeline

The Python package is custom-shoe-tree, the import module is custom_shoe_tree, and the CLI entry point is custom-shoe-tree.

Phase	Module	What it does	Main outputs
1. Input normalization and measurement	align.py, measure.py	Aligns the scan, detects units, fixes semantic orientation, snaps heel/sole coordinates, and extracts anatomical measurements.	out/phase1/<scan_id>/mesh_aligned.obj, measurements.json, audit.json, annotated.png
2. Template registration	template.py	Prepares the canonical template shoe tree, decimates it, measures it, and builds template landmarks.	out/phase2/base_shoe_tree_decimated.obj, base_shoe_tree_measurements.json, template_landmarks.json, template_landmarks.png
3. Parametric warping	warp.py	Samples 60 template sections, applies 7 width and 8 height adaptive control points, smooths the deformation, and trims collar fins.	out/phase3/<scan_id>/shoe_tree_warp.obj, warp_report.json, render.png
4. Geometry refinement	refine.py	Runs non-rigid ICP with sole protection and toe sock-wrap cleanup, then selects the safest refined mesh.	out/phase4/<scan_id>/shoe_tree_refined.obj, refine_report.json, pass renders, comparison.png
5. Final export	finalize.py	Repairs topology, exports fabrication-ready OBJ/STL files, and reports print-volume stats.	out/phase5/<scan_id>/shoe_tree_<scan_id>.obj, shoe_tree_<scan_id>.stl, pipeline_report.json, review.png
6. Print splitting (optional)	split.py	Cleans the STL, splits it along the Y-axis, and adds snap-fit tabs/sockets for print-bed fabrication.	out/phase6_split/<scan_id>/shoe_tree_<scan_id>_heel_tabs.stl, shoe_tree_<scan_id>_toe_sockets.stl, split_report.json

The template input file is intentionally still named cc_base_last.obj because that is the third-party source asset name. It is ignored by Git and expected at template/cc_base_last.obj. The optional print-splitting phase integrates the partner cleanup prototype from mesh_tools.py into the package; the simpler split_mesh.py prototype is superseded by this packaged flow.

Measured Parameters

The measurement stage extracts:

• Foot length
• Ball width and ball perimeter
• Toe box width, toe box height, and toe angle
• Arch peak, arch height, arch length, and arch type
• Heel center and heel width
• Max foot width and max height
• Adaptive ball position and longitudinal height profile

Fabrication Notes

The presentation prototype explored splitting the model along the Y-axis midpoint so larger shoe trees fit common printer beds. The split design uses parametric tabs and sockets so the halves can clip together after printing. The fabrication tests used BambuLab and Creality printers; the largest size 13 print took about 6.5 hours at 5% infill, while smaller split prints took about 3 hours.

The final slides show the printed shoe tree fitted into a Nike trainer. See the included presentation PDF for the result photos and fabrication context.

Repository Layout

.
|-- README.md
|-- pyproject.toml
|-- uv.lock
|-- src/custom_shoe_tree/
|   |-- cli.py
|   |-- align.py
|   |-- measure.py
|   |-- template.py
|   |-- warp.py
|   |-- refine.py
|   |-- finalize.py
|   |-- split.py
|   |-- viz.py
|   `-- io.py
|-- tests/
|-- scripts/smoke_all_scans.py
|-- sample-foot-scans/
|-- template/
|   `-- cc_base_last.obj       # local ignored asset
|-- out/
|-- webapp/
|   |-- backend/
|   `-- frontend/
`-- docs/
    |-- demo/
    `-- presentation/

*.obj and *.stl files are ignored, so cloned checkouts need local mesh assets before the full pipeline can run. Keep the template asset at template/cc_base_last.obj; local sample scans can be placed under sample-foot-scans/.

Quick Start: CLI

uv sync
uv run custom-shoe-tree --help
uv run custom-shoe-tree pipeline "sample-foot-scans/0014-B.obj"
uv run custom-shoe-tree pipeline --split-for-print "sample-foot-scans/0014-B.obj"
uv run custom-shoe-tree split "out/phase5/0014-B/shoe_tree_0014-B.stl"
uv run python scripts/smoke_all_scans.py
uv run pytest

The pipeline command requires both a foot-scan OBJ and the ignored template asset at template/cc_base_last.obj. If the sample scans are not present in your checkout, replace sample-foot-scans/0014-B.obj with the path to your own scan.

Quick Start: Web App

Start the backend:

cd webapp/backend
python3 -m venv .venv
source .venv/bin/activate
pip install -e ../..
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

Start the frontend in a second terminal:

cd webapp/frontend
npm install
npm run dev

Open http://localhost:5173, upload a .obj foot scan, review the measurements, adjust the shoe-tree allowance if needed, choose whether to split the STL for print-bed fabrication, generate the model, and download either the single STL or the split heel/toe STLs.

Tech Stack

• Python 3.12
• trimesh, scipy, numpy, shapely, networkx, manifold3d, fast-simplification
• pymeshfix and pyvista for Phase 6 split-prep cleanup
• FastAPI backend
• Vite + React frontend

Datasets And Related Work

The deck references FOOT3D / FIND as related 3D foot-scan work: https://www.ollieboyne.com/FIND.

This repository is structured for the FABRIC sample scans used during the project. Because mesh files are ignored, local scan files are expected under sample-foot-scans/ when running the smoke script.

Authors And Acknowledgments

Built by Alex, James, and Naijun as a Computational Fabrication (CASCS581) partner project.