3DAutoMod

AI-powered car customization studio. Upload a photo of a car, change its paint colour, swap its rims, and generate a 3D model — then send the finished spec to nearby mechanics for bids.

Final Year Project.

Demo

FYP Vid.mp4 — full walkthrough of the app (click the file, then View raw to play).

Features

• Studio — upload a car image, segment its parts, recolour the body, replace rims with a chosen rim image
• 3D generation — turn the customized 2D car into a viewable 3D model (@react-three/fiber + drei viewer)
• Mechanics — browse mechanics, filter, request bids on a customization job
• Bids & Chat — mechanics submit bids; owner chats with the mechanic in-app
• Auth — Supabase auth with owner / mechanic roles and row-level security

Tech Stack

Layer	Stack
Frontend	React 18, TypeScript, Vite, Tailwind CSS, Framer Motion, React Router
3D	Three.js, @react-three/fiber, @react-three/drei
Backend	Python, FastAPI, Uvicorn
Vision	Detectron2, OpenCV, COCO-format instance segmentation
Data / Auth	Supabase (Postgres, Auth, RLS)

Repository Layout

src/                     React frontend
  components/studio/     Upload zone, 3D viewer, toolbox, navigation
  components/mechanics/  Mechanic list, bid list, filters, selection modal
  components/chat/       Chat header, history, input
  components/auth/       Login / signup form
  pages/                 Home, Studio, Mechanics, MechanicBids, Chat
  lib/supabase.ts        Supabase client
APIs/                    FastAPI vision backend
  main.py                API entrypoint - mounts all endpoints
  colorChangeFastAPI_partseg.py    Colour change via part segmentation
  rim_change_coco_FastAPI.py       Rim replacement (COCO masks)
  segmentation_FastAPI.py          Car part segmentation
  API_file_return_3d.py            2D -> 3D model generation
  rims_change/rims/                Rim segmentation notebook + scripts
supabase/migrations/     Postgres schema, RLS policies, auth triggers

API Endpoints

Served by APIs/main.py on http://0.0.0.0:8001, interactive docs at /docs.

Method	Path	Body	Purpose
POST	/change_car_color	file, r, g, b	Recolour the car body
POST	/segment_car	file	Return part segmentation
POST	/segment_car_rims	car_file, rim_file	Swap rims onto the car
POST	/generate_3d	file	Generate a 3D model from the image

Setup

1. Frontend

npm install
cp .env.example .env      # fill in your Supabase project URL and anon key
npm run dev

2. Backend

cd APIs
python -m venv .venv && .venv\Scripts\activate    # Windows
pip install -r requirements.txt

Detectron2 must be installed separately or the build fails:

pip install --no-build-isolation "detectron2 @ git+https://github.com/facebookresearch/detectron2.git@9604f5995cc628619f0e4fd913453b4d7d61db3f"

Then run:

python main.py

3. Database

Apply supabase/migrations/ to your Supabase project. It creates the profiles table, enables RLS, and adds a trigger that populates a profile on signup.

Datasets and Model Weights

Not committed — they exceed GitHub's 100 MB per-file limit and would bloat the repo:

• APIs/rims_change.zip (114 MB) — rim segmentation dataset
• APIs/JPEGImages.zip (18 MB) — training images
• APIs/Car-Parts-Segmentation/ (110 MB) — third-party dataset, clone it from https://github.com/dsmlr/Car-Parts-Segmentation
• Trained .pth weights

Place them back in the paths above to run training or the full segmentation pipeline locally.

Environment Variables

Name	Description
VITE_SUPABASE_URL	Supabase project URL
VITE_SUPABASE_ANON_KEY	Supabase anon (public) key

.env is gitignored. Use .env.example as the template.