Emoji-Scope

Emoji-Scope 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 emoji images into different style categories using the SiglipForImageClassification architecture.

    Classification Report:
                     precision    recall  f1-score   support
    
        Apple Style     0.9336    0.8538    0.8919       725
       DoCoMo Style     0.9130    0.8400    0.8750       100
     Facebook Style     0.8713    0.8915    0.8813       691
        Gmail Style     0.8289    0.8750    0.8514       288
       Google Style     0.8725    0.9505    0.9098       727
    JoyPixels Style     0.8960    0.9614    0.9276       726
         KDDI Style     0.9444    0.9333    0.9389       255
      Samsung Style     0.9584    0.9681    0.9632       690
     SoftBank Style     0.8407    0.8053    0.8226       190
      Twitter Style     0.9939    0.8900    0.9390       727
      Windows Style     0.9949    0.9949    0.9949       583
    
           accuracy                         0.9200      5702
          macro avg     0.9134    0.9058    0.9087      5702
       weighted avg     0.9222    0.9200    0.9201      5702

The model categorizes images into eleven emoji styles:

• Class 0: "Apple Style"
• Class 1: "DoCoMo Style"
• Class 2: "Facebook Style"
• Class 3: "Gmail Style"
• Class 4: "Google Style"
• Class 5: "JoyPixels Style"
• Class 6: "KDDI Style"
• Class 7: "Samsung Style"
• Class 8: "SoftBank Style"
• Class 9: "Twitter Style"
• Class 10: "Windows Style"

Run with Transformers🤗

!pip install -q transformers torch pillow gradio

import gradio as gr
from transformers import AutoImageProcessor
from transformers import SiglipForImageClassification
from transformers.image_utils import load_image
from PIL import Image
import torch

# Load model and processor
model_name = "prithivMLmods/Emoji-Scope"
model = SiglipForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)

def emoji_classification(image):
    """Predicts the style category of an emoji 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()
    
    labels = {
        "0": "Apple Style", 
        "1": "DoCoMo Style", 
        "2": "Facebook Style", 
        "3": "Gmail Style", 
        "4": "Google Style", 
        "5": "JoyPixels Style", 
        "6": "KDDI Style", 
        "7": "Samsung Style", 
        "8": "SoftBank Style", 
        "9": "Twitter Style", 
        "10": "Windows Style"
    }
    predictions = {labels[str(i)]: round(probs[i], 3) for i in range(len(probs))}
    
    return predictions

# Create Gradio interface
iface = gr.Interface(
    fn=emoji_classification,
    inputs=gr.Image(type="numpy"),
    outputs=gr.Label(label="Prediction Scores"),
    title="Emoji Style Classification",
    description="Upload an emoji image to classify its style."
)

# Launch the app
if __name__ == "__main__":
    iface.launch()

Intended Use:

The Emoji-Scope model is designed to classify emoji images based on different style categories. Potential use cases include:

• Emoji Standardization: Identifying different emoji styles across platforms.
• User Experience Design: Helping developers ensure consistency in emoji usage.
• Digital Art & Design: Assisting artists in selecting preferred emoji styles.
• Educational Purposes: Teaching differences in emoji representation.