README.md

metadata

license: apache-2.0
datasets:
  - qwertyforce/scenery_watermarks
language:
  - en
base_model:
  - google/siglip2-base-patch16-224
pipeline_tag: image-classification
library_name: transformers
tags:
  - Image-Classification
  - Watermark-Detection
  - SigLIP2

Watermark-Detection-SigLIP2

Watermark-Detection-SigLIP2 is a vision-language encoder model fine-tuned from google/siglip2-base-patch16-224 for binary image classification. It is trained to detect whether an image contains a watermark or not, using the SiglipForImageClassification architecture.

Watermark detection works best with crisp and high-quality images. Noisy images are not recommended for validation.

SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features https://arxiv.org/pdf/2502.14786

Classification Report:
              precision    recall  f1-score   support

No Watermark     0.9290    0.9722    0.9501     12779
   Watermark     0.9622    0.9048    0.9326      9983

    accuracy                         0.9427     22762
   macro avg     0.9456    0.9385    0.9414     22762
weighted avg     0.9435    0.9427    0.9424     22762

---

Label Space: 2 Classes

The model classifies an image as either:

Class 0: "No Watermark"
Class 1: "Watermark"

---

Install dependencies

pip install -q transformers torch pillow gradio

---

Inference Code

import gradio as gr
from transformers import AutoImageProcessor, SiglipForImageClassification
from PIL import Image
import torch

# Load model and processor
model_name = "prithivMLmods/Watermark-Detection-SigLIP2"  # Update this if using a different path
model = SiglipForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)

# Label mapping
id2label = {
    "0": "No Watermark",
    "1": "Watermark"
}

def classify_watermark(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()
    
    prediction = {
        id2label[str(i)]: round(probs[i], 3) for i in range(len(probs))
    }

    return prediction

# Gradio Interface
iface = gr.Interface(
    fn=classify_watermark,
    inputs=gr.Image(type="numpy"),
    outputs=gr.Label(num_top_classes=2, label="Watermark Detection"),
    title="Watermark-Detection-SigLIP2",
    description="Upload an image to detect whether it contains a watermark."
)

if __name__ == "__main__":
    iface.launch()

---

Demo Inference

Watermark

No Watermark

Intended Use

Watermark-Detection-SigLIP2 is useful in scenarios such as:

• Content Moderation – Automatically detect watermarked content on image sharing platforms.
• Dataset Cleaning – Filter out watermarked images from training datasets.
• Copyright Enforcement – Monitor and flag usage of watermarked media.
• Digital Forensics – Support analysis of tampered or protected media assets.