SigLIP2 Content Filters - Models
Collection
Moderation, Balance, Classifiers
β’
9 items
β’
Updated
β’
2
Coral-Health is an image classification vision-language encoder model fine-tuned from google/siglip2-base-patch16-224 for a single-label classification task. It is designed to classify coral reef images into two health conditions using the SiglipForImageClassification architecture.
Classification Report:
precision recall f1-score support
Bleached Corals 0.8677 0.7561 0.8081 4850
Healthy Corals 0.7665 0.8742 0.8168 4442
accuracy 0.8125 9292
macro avg 0.8171 0.8151 0.8124 9292
weighted avg 0.8193 0.8125 0.8122 9292
The model categorizes images into two classes:
!pip install -q transformers torch pillow gradio
import gradio as gr
from transformers import AutoImageProcessor
from transformers import SiglipForImageClassification
from PIL import Image
import torch
# Load model and processor
model_name = "prithivMLmods/Coral-Health"
model = SiglipForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)
# Updated labels
labels = {
"0": "Bleached Corals",
"1": "Healthy Corals"
}
def coral_health_detection(image):
"""Predicts the health condition of coral reefs in the image."""
image = Image.fromarray(image).convert("RGB")
inputs = processor(images=image, return_tensors="pt")
with torch.no_grad():
outputs = model(**inputs)
logits = outputs.logits
probs = torch.nn.functional.softmax(logits, dim=1).squeeze().tolist()
predictions = {labels[str(i)]: round(probs[i], 3) for i in range(len(probs))}
return predictions
# Create Gradio interface
iface = gr.Interface(
fn=coral_health_detection,
inputs=gr.Image(type="numpy"),
outputs=gr.Label(label="Prediction Scores"),
title="Coral Health Detection",
description="Upload an image of coral reefs to classify their condition as Bleached or Healthy."
)
# Launch the app
if __name__ == "__main__":
iface.launch()
The Coral-Health model is designed to support marine conservation and environmental monitoring. Potential use cases include:
Base model
google/siglip2-base-patch16-224