This repository presents the training and evaluation of a YOLOv8s object detection model on a custom 21-class animal dataset, including species like Cattle, Goat, Sheep, Chicken, Duck, and more. The goal is accurate localization and classification of diverse animals in natural environments.
"Cattle", "Goat", "Sheep", "Chicken", "Duck", "Pig", "Horse", "Rabbit", "Tiger", "Leopard", "Fox", "Bear", "Snake", "Lizard", "Dog", "Eagle", "Mouse", "Monkey", "Porcupine", "Elephant", "Reptile"
- Number of Classes: 24 animal species
- Label Format: YOLO (normalized
[x_center, y_center, width, height]) - Annotation Distribution:
- Pairwise Feature Distribution (Correlogram):
- Framework: Ultralytics YOLOv8
- Model: YOLOv8s (Small variant)
- Training Duration: 50 epochs
- Losses Monitored: Bounding box loss, classification loss, distribution focal loss (DFL)
- Significant improvement in:
- Precision
- Recall
- mAP@0.5 and mAP@0.5:0.95
- Loss curves (box, cls, dfl) show consistent decrease, indicating healthy convergence.
| Raw Confusion Matrix | Normalized Confusion Matrix |
|---|---|
 |
 |
- Good separation among most classes.
- Minor confusion among similar species (e.g., Goat vs. Sheep).
Visual comparisons of ground truth and predictions:
| Ground Truth | Prediction |
|---|---|
 |
 |
 |
 |
 |
 |
- ✅ Trained on a balanced, multi-species animal dataset
- 📈 Achieved high accuracy and generalization
- 🔍 Minimal confusion between classes
- 🧠 Precise bounding box prediction across object sizes
- 🔄 Future-ready for YOLOv8m/l/x scaling or real-time deployment
- Address class imbalance for low-frequency species
- Train on YOLOv8m/l/x for improved accuracy
- Integrate real-time inference pipeline
- Experiment with augmentation and hyperparameter optimization