AlecKarfonta/cardcaptor-merged-dataset

CardCaptor Trading Card Dataset · merged_dataset_50x3

This is the dataset used to train the CardCaptor OBB Detector v3 model.

📦 Contents

• ~60,408 synthetic training renderings (dataset_0)
• ~220 real captures × 50× deterministic photometric multiplier ⇒ 11,000 effective handcrafted training images
• Validation: 55 handcrafted frames (210 ground-truth oriented boxes) to mirror real-world usage
• Format: YOLO OBB (class cx cy w h x1 y1 x2 y2 x3 y3 x4 y4) under images/ + labels/

➗ Directory Layout

merged_dataset_50x/
├── images/
│   ├── train/*.jpg
│   └── val/*.jpg
├── labels/
│   ├── train/*.txt
│   └── val/*.txt
├── merged_dataset.yaml
├── merged_dataset_visualization/
└── source_datasets_visualization/

🧪 Usage

git lfs install
git clone https://huggingface.co/datasets/AlecKarfonta/cardcaptor-merged-dataset
cd cardcaptor-merged-dataset

In Ultralytics:

from ultralytics import YOLO
model = YOLO("cardcaptor_v3_best.pt")
model.train(data="merged_dataset.yaml", epochs=50, imgsz=1088)

✍️ Notes

• Validation split favors hand-crafted images (≈90%) for realistic evaluation.
• Augmentation multiplier and deterministic photometric pre-augmentation create massive diversity without distorting labels.
• Visualization assets (raw + augmented + validation predictions) are included for quick inspection.