---
license: agpl-3.0
library_name: ultralytics
pipeline_tag: object-detection
base_model: Ultralytics/YOLO11
tags:
- object-detection
- ultralytics
- yolo11
- onnx
- onnxruntime
- casino-chips
- chip-detection
- casino
- live-game-defender
model-index:
- name: lgd-chips-gen1
  results:
  - task:
      type: object-detection
    dataset:
      type: lgd-chip-gate-blackjack
      name: LGD chip_test blackjack gate (private, never trained on)
    metrics:
    - type: recall
      value: 0.91
      name: Chip recall @50
    - type: precision
      value: 0.77
      name: Precision
---

# LGD Chips — Gen 1 · YOLO11s casino chip-colour detector (5 colours)

The first generation of the **Live Game Defender (LGD)** casino-chip detector: a **YOLO11s** object
detector that finds gaming chips on the felt and classifies each by **colour**. Trained on
point-labeled blackjack-table footage. Colours only — colour→denomination is per-casino
configuration downstream.

> **Superseded** by **[`lgd-chips-gen2`](https://huggingface.co/sroot/lgd-chips-gen2)** (the current
> served model — a dual-felt blend covering both blackjack and Ultimate Texas Hold'em, with a
> different venue-matched palette). Gen 1 is published for provenance.

## Generations

| Gen | Repo | Palette | Frozen-gate verdict | Status |
|---|---|---|---|---|
| 1 | **lgd-chips-gen1** (this) | black / white / green / red / pink | blackjack gate: recall 0.91 / precision 0.77 | superseded |
| 2 | [lgd-chips-gen2](https://huggingface.co/sroot/lgd-chips-gen2) | black / green / pink / purple / yellow | UTH gate: 0.80 / 0.85 / colour 0.97 | ✅ current |

Card detectors: [lgd-cards-gen1](https://huggingface.co/sroot/lgd-cards-gen1) · [lgd-cards-gen2](https://huggingface.co/sroot/lgd-cards-gen2) · [lgd-cards-gen3](https://huggingface.co/sroot/lgd-cards-gen3).

## Classes (5, in `model.classes.json` order)

```
chip_black  chip_white  chip_green  chip_red  chip_pink
```

## Files

- `model.onnx` — ONNX export (run with `onnxruntime`).
- `model.classes.json` — ordered class-name sidecar (index → colour).
- `metrics.json` — training config + evaluation.

## Training

- **Base:** Ultralytics `yolo11s.pt` (COCO-pretrained), 60 epochs, `imgsz=640`.
- **Data:** point-labeled chips from our blackjack-table proof-of-concept footage (labeled by
  vision agents — chip colour is a small class set, so no external labeling service was used).
- **Hardware:** NVIDIA RTX 3060 (12 GB).

## Metrics — frozen gate (never trained on)

| Gate | Recall | Precision |
|---|---|---|
| `chip_test` (blackjack) | 0.91 | 0.77 |

> ⚠️ **Not casino accuracy (rule of the project).** Gate footage is our own PoC recordings. This
> generation also produced **false `chip_black` boxes on printed felt markings** (fixed in gen 2).
> Dataset-internal `mAP@50` (0.97) in `metrics.json` is not real-world accuracy.

## Usage

```python
import json, onnxruntime as ort
sess = ort.InferenceSession("model.onnx", providers=["CPUExecutionProvider"])
colours = json.load(open("model.classes.json"))   # index -> "chip_black", ...
# Detect boxes per colour, then count per class for stack/bet estimation.
```

## License & provenance

**AGPL-3.0**, a fine-tune of **Ultralytics YOLO11** (`yolo11s.pt`, AGPL-3.0) — these weights inherit
AGPL-3.0 and **are not an original work of ours**. Networked deployment triggers AGPL §13 (offer the
Corresponding Source). `onnxruntime` (MIT) keeps the inference code AGPL-free; the weights stay AGPL.

Built for **Live Game Defender (LGD)** — an on-prem AI integrity monitor for live casino table
games. © 2026 TechTools s.r.o.