app.py

import torch
from transformers import ViTForImageClassification, ViTImageProcessor
import torch.nn.functional as F
from PIL import Image
import gradio as gr

model = ViTForImageClassification.from_pretrained('dj-dawgs-ipd/IPD-Image-ViT-Finetune')
processor = ViTImageProcessor.from_pretrained('google/vit-base-patch16-224')

class_names = ['cut_throat_gesture', 'finger_gun_to_the_head', 'middle_finger', 'slanted_eyes_gesture', 'swastika']

def predict(image):
    inputs = processor(images=image, return_tensors="pt")
    
    with torch.no_grad():
        outputs = model(**inputs).logits
    
    # predicted_class_idx = outputs.argmax(-1).item()
    # predicted_class = class_names[predicted_class_idx]
    
    # return predicted_class

    probabilities = F.softmax(outputs, dim=1)
    predicted_class_idx = probabilities.argmax(-1).item()
    predicted_class = class_names[predicted_class_idx]
    confidence_score = probabilities[0][predicted_class_idx].item()
    
    return predicted_class, confidence_score

iface = gr.Interface(fn=predict, 
                     inputs=gr.Image(type="pil"), 
                     outputs=[gr.Label(num_top_classes=1, label="Class"), gr.Label(label="Score")],
                     title="Hateful Content Detection",
                     description="Upload an image to classify hateful gestures or symbols")

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