from transformers import BlipProcessor, BlipForConditionalGeneration
from PIL import Image
import gradio as gr
import torch

# Load BLIP model and processor
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")
model.eval()
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)

# Inference function
def generate_caption(image):
    if image.mode != "RGB":
        image = image.convert("RGB")
    
    inputs = processor(image, return_tensors="pt").to(device, torch.float16)
    output = model.generate(**inputs, max_new_tokens=50)
    caption = processor.decode(output[0], skip_special_tokens=True)
    return caption

# Gradio interface
iface = gr.Interface(
    fn=generate_caption,
    inputs=gr.Image(type="pil"),
    outputs="text",
    title="Construction Site Image-to-Text Generator",
    description="Upload a site photo. The model will detect and describe construction activities."
)

iface.launch()