resist.

Identify any resistor instantly — upload a photo, use your camera, or pick bands manually.

A clean, minimal web app that decodes resistor color bands. Runs fully offline using trained ONNX models, with an optional Gemini API fallback.

---

How it works

Photo → YOLOv8n (detects band bounding boxes) → resistance formula → result

Model trained on Google Colab (free T4 GPU), exported to ONNX. Zero API costs. Runs on CPU.

---

Features

• AI photo detection — works in any lighting once models are trained
• Live camera — capture and analyze in one tap
• Manual band picker — live resistor SVG, 3/4/5/6-band support, updates in real time
• Auto band orientation — automatically flips reversed bands (gold/silver detected first)
• Correct resistance math — separate formulas per band count, tolerance and tempco display
• Offline-first — ONNX inference, no internet needed after setup
• Gemini override — toggle to use Gemini Vision AI instead of local models

---

Quick start

1. Clone and install

git clone https://github.com/praneel7015/resist
cd resist

python -m venv .venv
source .venv/bin/activate        # Windows: .\.venv\Scripts\Activate.ps1

pip install -r requirements.txt

2a. Run with Gemini (no training required)

Get a free key at aistudio.google.com:

cp .env.example .env
# Edit .env: GEMINI_API_KEY=AIza...
python app.py

2b. Run with local ONNX models (after training)

# Place trained models in:
# inference/models/band_detector.onnx
# inference/models/band_classes.json

python app.py   # auto-detects models and runs offline

You can also check the "Use Gemini AI" toggle in the UI to override local detection and use the Gemini API instead.

---

Training your own models

Overview

The detection pipeline uses YOLOv8n to find where each band is on the resistor body (object detection).

Step 1 — Get the Colab notebook

Open notebooks/resist_train.ipynb in Google Colab.

Or open directly:

Step 2 — Set runtime to T4 GPU

In Colab: Runtime → Change runtime type → T4 GPU

Step 3 — Get a Roboflow API key

1. Create a free account at roboflow.com
2. Profile icon → Settings → Roboflow API → copy your key
3. Paste it into the notebook cell

Dataset used: Resistor Band Detection — annotated resistor images with per-band bounding boxes and color labels.

Step 4 — Run all cells

The notebook (~25 minutes on T4) will:

1. Download and inspect the dataset
2. Train YOLOv8n for color detection
3. Export model to ONNX
4. Save files to your Google Drive

Step 5 — Download and deploy

Copy from Google Drive → resist_models/ into your project:

resist/
└── inference/
    └── models/
        ├── band_detector.onnx       (~11 MB)
        └── band_classes.json

Restart Flask — it will auto-detect the models and switch to offline mode.

---

Project structure

resist/
├── app.py                    # Flask server — auto-selects local vs Gemini
├── requirements.txt
├── Dockerfile
├── .env.example
├── notebooks/
│   └── resist_train.ipynb    # Full Colab training notebook
├── inference/
│   ├── detector.py           # ONNX inference — YOLOv8 pipeline
│   └── models/               # Put your .onnx files here (not in git)
├── templates/
│   └── index.html
└── static/
    ├── style.css
    └── app.js

---

Environment variables

Variable	Required	Description
GEMINI_API_KEY	Only without local models	Free key from aistudio.google.com
PORT	No	Server port (default 8080 in Docker, 5000 local)

---

Deployment

Docker

docker build -t resist .
docker run -p 8080:8080 resist
# Add -e GEMINI_API_KEY=... if using Gemini fallback

AWS App Runner

App Runner scales near-zero when idle.

1. Push to GitHub
2. AWS Console → App Runner → Create service
3. Source: Repository → connect GitHub → select this repo
4. Build: Dockerfile (auto-detected)
5. Port: 8080
6. Environment variable: GEMINI_API_KEY (only needed without local models)
7. Health check: /health
8. Deploy — auto-redeploys on every push to main

AWS Lambda + API Gateway (true serverless)

Add mangum to requirements.txt, then:

# bottom of app.py
from mangum import Mangum
handler = Mangum(app)

Roadmap

• Manual band picker (3–6 bands, live SVG preview)
• Camera capture mode
• Gemini Vision API integration
• YOLOv8 pipeline (Colab notebook)
• ONNX export for zero-dependency inference
• Auto band orientation (flips reversed gold/silver)
• Gemini override toggle in UI
• Ohm's law calculator
• LED resistor calculator (supply voltage + LED specs → resistor value)
• Voltage divider calculator
• Series / parallel resistance calculator
• E12 / E24 / E96 nearest standard value finder
• SMD resistor code decoder
• Scan history

---

Tips for better detection

• Good even lighting — avoid shadows across the bands
• Plain background (white or black) behind the resistor
• Get close so the resistor body fills most of the frame
• If confidence shows "low", try the Manual tab or toggle on Gemini AI

---

Contributing

Pull requests welcome. If you've improved the training pipeline or have a better dataset, please open an issue first.

---

License

MIT — see LICENSE