Multicrab Detector Yolo11s100

Multicrab_detector_yolo11s100

A YOLO11s object detection model fine-tuned to detect and classify multiple crab species in a single frame, including scenes with several individuals or overlapping crabs. Trained for 100 epochs.

Model Details

• Base architecture: YOLO11s (Ultralytics), fine-tuned from yolo11s.pt
• Task: Object detection (bounding boxes)
• Classes: Multiple crab species (full class list to be added)
• Epochs: 100
• Input image size: 640x640
• Framework: PyTorch / Ultralytics
• License: AGPL-3.0 (Ultralytics default — update if the dataset license requires otherwise)

Intended Use

This model detects and classifies multiple crab species in a single frame, including scenes with several individuals or overlapping crabs. Suitable for applications such as underwater/field crab monitoring, species-count automation, and similar object-detection workflows.

Training Data

• Dataset: "Crab_project" v1, exported via Roboflow on 2026-04-16. 1,017 source images, expanded to 3 versions per image via augmentation.
• Split: 897 train / 80 validation / 40 test images
• Source: Roboflow-managed annotation project ("Crab-fYh8"), annotated in YOLOv11 format
• Annotation format: YOLO (YOLOv11 export format from Roboflow)
• Preprocessing applied: Auto-orientation of pixel data (EXIF-orientation stripping)
• Augmentations applied (3x per source image):
  • 50% probability horizontal flip
  • 50% probability vertical flip
  • Random rotation: -14° to +14°
  • Random shear: -11° to +11° (horizontal), -12° to +12° (vertical)
  • Random brightness adjustment: -18% to +18%
  • Random Gaussian blur: 0 to 4.9 pixels
  • Salt-and-pepper noise applied to 1.72% of pixels

Training Procedure

Model: yolo11s.pt (pretrained weights, Ultralytics)
Epochs: 100
Image size: 640x640
Optimizer: Ultralytics default (SGD)
Learning rate (lr0): 0.001
Patience (early stopping): 10

Evaluation Results

Evaluated on: 40-image held-out test split from the Roboflow "Crab_project" v1 dataset. Quantitative metrics (mAP50, mAP50-95, precision, recall).

How to Use

from ultralytics import YOLO
# Load the model
model = YOLO("path/to/best.pt")  # or load directly from this HF repo
# Run inference on an image
results = model("path/to/image.jpg")
# Visualize / save results
results[0].show()
results[0].save("output.jpg")
# Run inference on a video / ROV footage stream
results = model("path/to/video.mp4", stream=True)
for r in results:
    boxes = r.boxes
    print(boxes.xyxy, boxes.conf, boxes.cls)

To load weights directly from the Hugging Face Hub:

from huggingface_hub import hf_hub_download
from ultralytics import YOLO
weights_path = hf_hub_download(repo_id="your-username/Multicrab_detector_yolo11s100", filename="best.pt")
model = YOLO(weights_path)

Citation

If you use this model, please cite:

@misc{multicrab_yolo11s_2026,
  title  = {Multicrab_detector_yolo11s100},
  author = {Mariam Essam},
  year   = {2026},
  howpublished = {Hugging Face Hub},
  url    = {https://huggingface.co/MariamEssam204/Multicrab_detector_yolo11s100}
}

Acknowledgements

Built on Ultralytics YOLO11. Dataset annotated and exported via Roboflow.