netrunner-card-identifier

Netrunner Card Identifier

A Chrome extension that identifies Netrunner cards in tournament and stream footage and shows the full card art and text inline. Inspired by Ocular Automaton (MTG) and NetReady Eyes (Netrunner desktop scanner).

What it does

• Alt+drag on a video to identify a card by its image. Works well on webcam-on-mat physical stream footage (Neon Static, tournament VODs, etc.) across all four card orientations.
• Hover over any embedded NetrunnerDB card image on any page to see the full card.

Card data and images come from NetrunnerDB; identification uses the Milo ONNX card-embedder running locally via WebAssembly.

Install

From the Chrome Web Store

(Pending review — link will be added once published.)

Unpacked (build from source)

Requires Python 3.10+ and bash (macOS/Linux; Windows works via Git Bash).

# 1. Set up the Python environment for the catalog builder
python3 -m venv .venv
source .venv/bin/activate
pip install -r tools/requirements.txt

# 2. Build the card embedding catalog (one-time, ~5 minutes)
python tools/build-catalog.py

# 3. Fetch the onnxruntime-web vendor files (~32 MB, one-time)
bash tools/fetch-vendor.sh

Then in Chrome:

1. chrome://extensions → enable Developer mode
2. Load unpacked → select this repo’s root directory
3. Pin the extension to your toolbar if you want quick access to the popup.

Use

Identify a card in a video (works on physical-cardboard footage — see Scope below):

1. Pause a YouTube/Twitch video showing a physical Netrunner game on a clear frame.
2. Hold ⌥ (Option) / Alt and drag a rectangle around the card. The selection rectangle locks to card aspect, in the orientation matching your drag direction.
3. The matched card image floats next to your cursor. Click it or press Escape to dismiss.

Note: Don’t test on jinteki.net video — that’s a known weak case and a separate recognizer is planned for a future release. Use webcam-on-mat tournament footage instead.

The page console (Cmd+Opt+I → Console) prints [netrunner-video] lines with the top-3 cosine matches per drag, useful when accuracy is off.

Hover-on-image:

• On NetrunnerDB, in a forum/blog post embedding card images, or anywhere else with NetrunnerDB-hosted images, hover any small card image to see the full art floating beside it. Suppressed automatically when the source image is already large enough to read.

Scope and known limitations

Use case	Works?
Physical cardboard play on webcam (Neon Static-style streams)	✓ Primary use case
Tournament webcam VODs	✓
Hover on NetrunnerDB / jinteki.net / forum card images	✓
jinteki.net play captured in YouTube/Twitch video	~ Experimental — see “jinteki augmentation” below
Cards smaller than ~50 px on a 1080p frame	✗ Not enough pixels after the embedder’s 448 × 448 downscale

The catalog ships with every Netrunner card NetrunnerDB tracks (the “Eternal” pool — FFG era through Null Signal era, ~2000 cards). Each card×printing is embedded at 4 rotations plus a “J” variant per rotation that simulates how the card looks in jinteki.net stream footage (top-cropped + downscaled). Total ≈19,384 catalog rows. To rebuild after a new set drops: python tools/build-catalog.py. To scope to just the current Standard format: python tools/build-catalog.py --pool standard.

jinteki augmentation (experimental)

The JP/JL/JU/JR catalog variants are the extension’s attempt to identify cards shown on YouTube/Twitch videos of jinteki.net play. They simulate jinteki’s installed-card appearance (top-cropped, downscaled). When the diagnostic top-3 log shows a winning orient=J*, that’s the augmented path firing.

Caveat: ICE cards share heavy visual layout (subroutines + strength box + faction symbol) and their J-variants score very close to each other. On a real jinteki video, ICE identification may often land in the “ambiguous” gate and refuse to commit. That’s the conservative default — better silent than confidently wrong. Cardboard webcam ICE remains the strong case.

Privacy

The extension makes no telemetry calls and stores nothing about you. It:

• Queries NetrunnerDB once per week to refresh card metadata.
• Downloads the Milo model from Hugging Face once, when you first build the catalog locally. All identification runs on your machine.
• Stores the card catalog and last-sync timestamp in chrome.storage.local.

Full privacy policy: store/privacy-policy.md.

Project layout

manifest.json
src/
  background/           service worker (NRDB sync, offscreen lifecycle)
  content/              overlay.js, hover.js, video-click.js
  lib/                  NRDB client, chrome.storage wrapper
  offscreen/            ML inference host (loads ONNX, runs nearest-neighbor)
  model/                embedder.onnx + catalog.bin + catalog.json (committed)
  vendor/ort/           onnxruntime-web (not committed; fetched by setup)
  icons/                extension icons (16/32/48/128)
  popup/                toolbar popup (sync status, "About" link)
tools/
  build-catalog.py      offline catalog builder (Python + onnxruntime)
  fetch-vendor.sh       downloads onnxruntime-web into src/vendor/ort/
  make-icons.py         regenerates src/icons/*.png from scratch
store/                  Chrome Web Store assets, listing copy, privacy policy

Credits

• NetReady Eyes by @eheiden — the Python desktop card scanner that pioneered the approach this extension borrows (Milo embedder + NRDB catalog + nearest-neighbor). Different form factor, same technical lineage.
• Milo by @HanClinto — the MobileViT-XXS card embedder that does the heavy lifting.
• NetrunnerDB — card data and images.
• Null Signal Games — Netrunner.

License

AGPL-3.0. Required because the bundled Milo model is AGPL-3.0. A permissively-licensed fork is possible if you swap the embedder.

This site is open source. Improve this page.