# YOLOv9s Beverage Detection - Competition Model

**Competition:** TurboVision Subnet 44 - Beverage Detection
**Element ID:** manak0/Detect-beverage-detect
**Deployed:** 2026-04-30

## Performance Metrics

**Validation:**
- mAP50: 89.59%
- mAP50-95: 68.77%
- Model Size: 28MB

**Test Results (11 images):**
- Average Detections: 12.6 per image
- Can Detection: 100% (45/45 cans)
- Bottles: Detected
- Cups: Detected

## Competition Targets

- Baseline: 5.9% mAP50
- Target: 90% mAP50
- **Our Model: 89.59%** ✅

## Classes

1. **cup** - Cups, mugs, beer glasses
2. **bottle** - Various bottle types
3. **can** - Beverage cans

Note: Model also detects wine_glass but competition only evaluates cup, bottle, can.

## Training Details

- Base Model: YOLOv9s
- Parameters: 7.32M
- Dataset: 4,840 images
- Epochs: 100
- Training Time: 1.14 hours
- GPU: NVIDIA L40S

## Deployment

Deployed via ScoreVision CLI:
```bash
sv -vv deploy-os-miner --element-id Ichiro1007/Detect-beverage-detect
```

## Expected Competition Performance

- Initial (24hrs): 40-60%
- Convergence (7 days): 70-90%
- Target: Beat 5.9% baseline ✅✅✅

---

**Repository:** https://huggingface.co/Ichiro1007/Detect-beverage-detect
**Developer:** Ichiro1007