README.md

# Training Logs - EfficientNet-B3
## Experiment: 20251111_031508

### Model Architecture
- **Base Model**: EfficientNet-B3 (pretrained on ImageNet)
- **Task**: Multi-task learning (Age Regression + Gender Classification)
- **Input Size**: 256x256

### Training Configuration
- **Batch Size**: 16
- **Learning Rate**: 0.0002
- **Optimizer**: AdamW (weight_decay=1e-4)
- **Scheduler**: CosineAnnealingWarmRestarts
- **Epochs Trained**: 34 / 40

### Best Results
- **Validation Harmonic Mean**: 0.9699
- **Validation Age RMSE**: 0.1175 years
- **Validation Gender F1**: 0.9469
- **Best Epoch**: 12

### Final Metrics (Last Epoch)
- **Val Loss**: 0.2847
- **Val Harmonic Mean**: 0.9650
- **Val Age RMSE**: 0.1177
- **Val Age MAE**: 0.0875
- **Val Gender F1**: 0.9375

### Files in this Dataset
- `pytorch_lightning/metrics_20251111_031508.csv` - Complete training history
- `summaries/training_summary_20251111_031508.json` - Detailed metrics and config
- `trackio/trackio_logs_20251111_031508.zip` - TrackIO logs (if available)

### Next Steps
Use the best checkpoint for inference:
```python
checkpoint_path = "checkpoints/efficientnet/effnet-epoch=12-val_harmonic_mean=0.9699.ckpt"
```