app.py

import gradio as gr
from transformers import VisionEncoderDecoderModel, ViTImageProcessor, AutoTokenizer
from PIL import Image
import torch


model_name = "nlpconnect/vit-gpt2-image-captioning"

print("Loading model components...")
model = VisionEncoderDecoderModel.from_pretrained(model_name)
feature_extractor = ViTImageProcessor.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)
print("Model loaded!")

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

def generate_caption(image):
    """
    Takes a PIL Image and returns a text caption.
    """
    if image is None:
        return "Please upload an image."
    
    # Ensure image is in RGB mode so it has 3 channels
    if image.mode != "RGB":
        image = image.convert(mode="RGB")
    
    # Preprocess the image
    pixel_values = feature_extractor(images=[image], return_tensors="pt").pixel_values
    pixel_values = pixel_values.to(device)
    
    # Generate output
    output_ids = model.generate(pixel_values, max_length=16, num_beams=4)
    
    # Decode text
    preds = tokenizer.batch_decode(output_ids, skip_special_tokens=True)
    return preds[0].strip()

demo = gr.Interface(
    fn=generate_caption,
    inputs=gr.Image(type="pil", label="Upload an Image"),
    outputs=gr.Textbox(label="AI Caption"),
    title="AI Image Captioner",
    description="Upload any photo, and the AI will describe what it sees using a Vision Transformer + GPT-2 model!",
    examples=[],
    theme="default"
)

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