DeepShark-ViT-Hatem-V1

🦈 vit-shark-model: Vision Transformer for Shark Species Classification

A fine-tuned Vision Transformer (ViT-B/16) model trained to classify 14 different shark species using image data.
This model leverages the power of transfer learning from the Hugging Face google/vit-base-patch16-224-in21k backbone.

---

📊 Model Summary

---

🐋 Classes (14 Total)

• 'blacktip_Shark'
• 'blue_shark'
• 'bull_Shark'
• 'great_White_Shark'
• 'grey_reef_shark'
• 'hammerhead_Shark'
• 'mako_Shark'
• 'sand_tiger_shark'
• 'sevengill_shark'
• 'silky_shark'
• 'silvertip_shark'
• 'tiger_Shark'
• 'whale_Shark'
• 'whitetip_Shark'

---

🧾 Dataset Summary

Dataset was preprocessed using ViTImageProcessor with resizing, normalization, and label mapping.

---

📈 Training Metrics

---

📊 Training Curves

Below are the plots showing the loss and accuracy over training epochs.

✅ Use Cases

Marine biology research

Wildlife conservation

Shark detection for underwater robotics

Educational tools for biodiversity studies

---

🚫 Limitations

Model may misclassify species with similar appearances.

Performance could degrade on blurry, low-resolution, or occluded images.

Fine-tuned on a relatively small dataset (~2000 images total).

---

(How to Use)

from transformers import ViTForImageClassification, ViTImageProcessor
from PIL import Image as PILImage
import torch

# المسار الذي تم حفظ النموذج فيه
model_path = "./vit-shark-model"

# تحميل المعالج والنموذج
image_processor = ViTImageProcessor.from_pretrained(model_path)
model = ViTForImageClassification.from_pretrained(model_path)

# (اختياري) نقل النموذج لوحدة المعالجة الرسومية (GPU)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
model.eval() # وضع التقييم

# مثال: معالجة صورة وتوقعها
image = PILImage.open("path/to/your/shark_image.jpg").convert("RGB")
inputs = image_processor(images=image, return_tensors="pt").to(device)

with torch.no_grad():
    outputs = model(**inputs)

logits = outputs.logits
predicted_class_idx = logits.argmax(-1).item()
predicted_label = model.config.id2label[predicted_class_idx]

print(f"Predicted class: {predicted_label}")

!pip install gradio transformers torch torchvision

import gradio as gr
from transformers import pipeline
import torch
from torchvision import transforms
from PIL import Image

# Load the model
try:
    classifier = pipeline("image-classification", model="HatemMoushir/DeepShark-ViT-Hatem-V1")
except Exception as e:
    print(f"Error loading model: {e}")
    print("Please ensure you have access to the model or that the model name is correct.")
    # Fallback or exit if model can't be loaded
    exit()

# Define a function to make predictions
def predict_shark_species(image):
    if image is None:
        return "Please upload an image."

    # The pipeline handles preprocessing, but sometimes explicit conversion helps
    # Ensure the image is in RGB format if it's not already
    if image.mode != 'RGB':
        image = image.convert('RGB')

    # Make prediction
    # The output is a list of dictionaries, e.g., [{'score': 0.99, 'label': 'Great White Shark'}]
    predictions = classifier(image)

    # Format the output
    if predictions:
        # Get the top prediction
        top_prediction = predictions[0]
        label = top_prediction['label'].replace('_', ' ').title() # Format label nicely
        score = top_prediction['score'] * 100 # Convert to percentage

        if score < 50:
           return "Prediction: Unknown (Low confidence)"
        else:
        return f"Prediction: **{label}**\nConfidence: **{score:.2f}%**"
    else:
        return "No prediction could be made."

# Create the Gradio interface
iface = gr.Interface(
    fn=predict_shark_species,
    inputs=gr.Image(type="pil", label="Upload Shark Image"),
    outputs="markdown",
    title="🦈 DeepShark-ViT-Hatem-V1: Shark Species Classifier",
    description="Upload an image of a shark to get a prediction of its species using the HatemMoushir/DeepShark-ViT-Hatem-V1 model.",
    examples=[
        # You can add example image paths here if you have them locally
        # e.g., ["path/to/your/shark_example1.jpg"]
    ]
)

# Launch the interface
if __name__ == "__main__":
    print("Starting Gradio interface...")
    iface.launch(share=True) # Set share=True to get a public link (useful for sharing)

---

(Acknowledgements)

Special thanks to the scientific community and the developers of the Hugging Face Transformers and PyTorch libraries for providing the essential tools needed to build and train this model. We also extend our sincere gratitude for the support and guidance provided by ChatGPT and Gemini throughout various stages of this project, which significantly contributed to its success. Finally, we thank the Colab platform for offering a suitable environment for training the model.

---

license:

cc-by-nc-sa-4.0

Author

Hatem Moushir [email protected]