yolov8-djahit

Model Card: YOLOv8 Damage & Stain Detection Model Overview

Model Name: yolov8-djahit Author / Hugging Face Username: aannugra Model Type: Object Detection (YOLOv8) Task: Detect clothing defects: outside tear, stain, tear, hole, normal

Description: This YOLOv8 model is trained on Amazon Rekognition datasets for garment damage detection. The model predicts bounding boxes for five classes of clothing defects:

outside_tear

stain

tear

hole

normal

It is intended for automated inspection of clothing images to help streamline repair or quality assurance workflows.

Dataset

Dataset Source: Amazon Rekognition custom labels Dataset Format: JSON manifest files converted to YOLOv8 format Number of Classes: 5 Train/Validation Split: Defined in data.yaml

Model Usage Inference from ultralytics import YOLO

Load model from Hugging Face

model = YOLO("aannugra/yolov8-djahit")

Run inference on a single image

results = model.predict(source="test.jpg", device="cpu")

Display results

results.show()

Batch Prediction

Predict on a folder of images

results = model.predict(source="images_folder", device="cpu") results.show()

Access Bounding Boxes for r in results: print(r.boxes.xyxy) # bounding box coordinates print(r.boxes.cls) # predicted class indices print(r.boxes.conf) # confidence scores

Training Details

Framework: Ultralytics YOLOv8

Hardware: Trained on NVIDIA RTX 3050 GPU

Optimizer: Adam (default YOLOv8)

Image Size: 640x640

Epochs: 50

Data Augmentation: Default YOLO augmentations (flip, mosaic, etc.)

Model Statistics

Model Size: ~14 MB (YOLOv8n)

Number of Classes: 5

Detection Type: Bounding Box (Object Detection)

Citation

If you use this model in your work, please cite it as:

@misc{aannugra2025yolov8djahit, author = {Andika Satya Nugra}, title = {YOLOv8 Damage & Stain Detection}, year = {2025}, howpublished = {\url{https://huggingface.co/aannugra/yolov8-djahit}} }

Example Images / Visuals

You can optionally add example images with predictions: