This YOLO12X model is fine-tuned on a custom dataset consisting of annotated aerial images for detecting two object classes: UAP and UAI. The training was conducted using the Ultralytics framework.

📁 Dataset

Images: Custom drone footage frames
Classes:
- 0: UAP
- 1: UAI
Annotation format: YOLO (TXT with class x_center y_center width height)
Split: No separate val/test — trained and evaluated on the full dataset

🧠 Model Details

Base: YOLOv12X
Params: 4.8M
Trained for: 100 epochs
Image size: 640×640
Batch size: 16
Optimizer: SGD (default Ultralytics settings)
Loss Function: CIoU + BCE
Augmentations: mosaic, random affine, color jitter, horizontal flip

📊 Evaluation Results

Aggregate Metrics

Metric	Value
mAP@0.5	0.995
mAP@0.5:0.95	0.983
box_loss	0.2368
cls_loss	0.207
dfl_loss	0.8056

🖼️ Sample Predictions

🚀 Inference Example

python

from ultralytics import YOLO

model = YOLO("momererkoc/uap-uai-object-detection-yolo12x")
results = model("path/to/image.jpg")
results[0].show()  # show prediction