sroot/lgd-cards-video-day1

LGD Cards — Day-1 PoC Video (YOLO card-pip dataset)

Auto-labeled playing-card corner-pip detection tiles from the first day of our proof-of-concept table recordings, for Live Game Defender (LGD) — an on-prem AI integrity monitor for live casino table games. This is the day-1 training set behind the lgd-cards-gen2 detector.

• Format: YOLO — images/{train,val} + labels/{train,val}, data.yaml (52 classes, rank+suit).
• ~4,630 tiles, serve-matching 2×2 tiling of the source frames.
• Provenance: boxes.jsonl / manifest.json record every box's origin. Labels were proposed by a card detector and verified/named by an LLM over a closed 52-code vocabulary (labeling cost ≈ $3.41). QA review crops are intentionally not included.

Classes (52)

10C 10D 10H 10S 2C 2D 2H 2S 3C 3D 3H 3S 4C 4D 4H 4S 5C 5D 5H 5S 6C…9S AC AD AH AS JC JD JH JS KC KD KH KS QC QD QH QS (C=Clubs D=Diamonds H=Hearts S=Spades). The detector boxes pips, ~2 per card.

⚠️ Honesty note

These are our own proof-of-concept recordings, not live casino footage, and the labels are LLM-verified, not fully human-verified — treat this as a mechanism/reproducibility set, not a casino-accuracy ground truth. Card models mixed this with the Public-Domain Roboflow augmented-startups/playing-cards-ow27d v4 base set (not redistributed here — get it from Roboflow).

Family

Datasets: day-1 (this) · day-2 · day-3 · holdout. Models: lgd-cards-gen1/2/3, lgd-chips-gen1/2.

License CC-BY-NC-4.0 (non-commercial, attribution). © 2026 TechTools s.r.o.