ScrubTechAR

An XR software that allows surgeons to keep track of surgical tools during/before/after an OR, to reduce up to 90 minutes per surgery.

---

Surgical Instruments Detected

Class ID	Instrument	Training images
0	Scalpel	550
1	Straight Dissection Clamp	460
2	Straight Mayo Scissor	450
3	Curved Mayo Scissor	550

---

Setup

pip install -r requirements.txt

---

Workflow

1 — Build the dataset

Matches images in data/ with YOLO labels in Labels/label object names/, then splits everything into dataset/images/{train,val,test} and dataset/labels/{train,val,test}:

python setup_dataset.py
# Optional flags:
#   --train 0.80  --val 0.10  --test 0.10   (default 80/10/10 split)
#   --seed  42
#   --overwrite                              (rebuild from scratch)

2 — Train YOLOv11

python train.py
# Optional flags:
#   --model yolo11n.pt   (nano, default)
#   --model yolo11s.pt   (small)
#   --model yolo11m.pt   (medium — best accuracy/speed trade-off)
#   --epochs 100         (default)
#   --batch  16          (default)
#   --imgsz  640         (default)

Best weights are saved to runs/detect/scrubtech_v1/weights/best.pt.

3 — Count instruments

# Single image
python count_instruments.py --source data/Scalpel/bisturi1.jpg

# Entire folder
python count_instruments.py --source data/

# Save annotated output
python count_instruments.py --source data/ --save

# Live webcam
python count_instruments.py --source 0

# Custom weights
python count_instruments.py --source <path> --weights <path/to/best.pt>

Example output:

══════════════════════════════════════════════════
  Instrument count — data/
──────────────────────────────────────────────────
  Scalpel                            550  ██████████████████████
  Straight Dissection Clamp          460  ████████████████████
  Straight Mayo Scissor              450  ███████████████████
  Curved Mayo Scissor                550  ██████████████████████
──────────────────────────────────────────────────
  TOTAL                             2010
══════════════════════════════════════════════════

---

Project Structure

ScrubTechAR/
├── data/
│   ├── Scalpel/                    550 images (bisturi*.jpg)
│   ├── Curved Mayo Scissor/        550 images (tesouracurva*.jpg)
│   ├── Straight Dissection Clamp/  460 images (pinca*.jpg)
│   └── Straight Mayo Scissor/      450 images (tesourareta*.jpg)
├── Labels/
│   ├── label object names/         YOLO annotations — 4-class instrument ID
│   └── label top-bottom/           YOLO annotations — 2-class orientation
├── dataset/                        ← created by setup_dataset.py
│   ├── images/{train,val,test}/
│   └── labels/{train,val,test}/
├── runs/                           ← created by train.py / count_instruments.py
│   └── detect/scrubtech_v1/weights/best.pt
├── dataset.yaml                    YOLO dataset config
├── setup_dataset.py                Dataset preparation script
├── train.py                        YOLOv11 training script
├── count_instruments.py            Inference + instrument counting
└── requirements.txt