YOLO11 Pill Detection Model

Model Description

This model is a custom-trained YOLO11 object detection model developed for detecting full pills in images. It was trained using a labeled dataset prepared in Roboflow and exported in YOLO format.

The model identifies pills by predicting:

• Bounding boxes
• Confidence scores
• Object class labels

Currently, the model supports three detection classes:

• pill
• capsule
• tablet

This project is designed for pharmaceutical object detection research and computer vision experimentation.

---

Model Details

Model Type

• Architecture: YOLO11
• Task: Object Detection
• Framework: Ultralytics YOLO

Classes

names:
  0: pill
  1: capsule
  2: tablet

---

Intended Uses

Primary Use Cases

This model is intended for:

• Pill detection in images
• Pharmaceutical automation experiments
• Computer vision prototyping
• Medication localization in images/video

Out-of-Scope Use

This model is not intended for:

• Medical diagnosis
• Drug verification in clinical workflows
• Safety-critical pharmaceutical decisions

Predictions may be inaccurate under challenging imaging conditions.

---

Training Data

The model was trained on a custom Roboflow dataset containing images of Labeled_full_pill Computer Vision Model annotated with bounding boxes.

Dataset Characteristics

• Annotated in Roboflow
• Exported in YOLO8 format
• Single object class: pill

Dataset Split

Example:

• Train: 70%
• Validation: 20%
• Test: 10%

---

Training Procedure

The model was trained using Ultralytics YOLO11 with pretrained weights.

Training Hyperparameters

model: yolo11x.pt
epochs: 100
imgsz: 640
batch: 16
optimizer: auto
device: 6

Training Command

yolo train device=3 \
  model=ul://ultralytics/yolo11/yolo11x \
  data=ul://wijai-thongsom/datasets/labeled-full-pillv2iyolov8 \
  roject=wijai-thongsom/jolly-husky \
  name=yolo11x
  epochs=100 \
  imgsz=640 \
  batch=-1

---

Evaluation Results

Model performance was evaluated on the validation set using standard object detection metrics.

Metrics

Metric	Value
Precision	0.953978
Recall	0.932336
mAP50	0.965024
mAP50-95	0.728589

Replace these values with the actual metrics from your training results.

---

Inference

CLI Example

yolo detect predict \
  model=yolo11x.pt \
  source=image.jpg

Python Example

from ultralytics import YOLO

model = YOLO("yolo11x.pt")
results = model("image.jpg")

for result in results:
    print(result.boxes)

---

Limitations

The model performance may degrade in cases such as:

• Poor lighting
• Motion blur
• Partial occlusion
• Overlapping pills
• Pill appearances not represented in the training dataset

Performance is dependent on image quality and dataset diversity.

---

Bias and Risks

Because this model was trained on a custom dataset, its predictions may be biased toward:

• Specific pill colors
• Particular lighting conditions
• Limited pill shapes and sizes
• Background styles present in training data

Use caution when applying the model to images outside the training distribution.

---

Environmental Impact

Training object detection models requires computational resources that consume energy.

Training setup example:

• Hardware: GPU
• Framework: Ultralytics YOLO11
• Epochs: 100

For reproducibility, document:

• GPU type
• Training duration
• Energy consumption estimate

---

Model Files

Typical files included in this repository:

.
├── README.md
├── yolo11n.onnx
├── yolo11m.onnx
├── yolo11x.onnx
├── data.yaml
└── results.png

---

Citation

If you use this model, please cite:

@misc{yolo11-pill-detection,
  title={YOLO11 Pill Detection Model},
  author={Wijai Thongsom},
  year={2026},
  publisher={Hugging Face}
}

---

License

This model is released under the MIT License.

---

Acknowledgments

• Ultralytics for YOLO11
• Roboflow for dataset annotation/export
• Hugging Face Hub for model hosting