app.py

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

processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")

def generate_caption(image, caption_type, text):
    raw_image = Image.fromarray(image.astype('uint8'), 'RGB')
    
    if caption_type == "Conditional":
        caption = conditional_image_captioning(raw_image, text)
    else:
        caption = unconditional_image_captioning(raw_image)
    
    return caption

def conditional_image_captioning(raw_image, text):
    inputs = processor(raw_image, text, return_tensors="pt")
    out = model.generate(**inputs)
    caption = processor.decode(out[0], skip_special_tokens=True)
    return caption

def unconditional_image_captioning(raw_image):
    inputs = processor(raw_image, return_tensors="pt")
    out = model.generate(**inputs)
    caption = processor.decode(out[0], skip_special_tokens=True)
    return caption

input_image = gr.inputs.Image(label="Upload an Image")
input_text = gr.inputs.Textbox(label="Enter Text (for Conditional Captioning)")
radio_button = gr.inputs.Radio(choices=["Conditional", "Unconditional"], label="Captioning Type")
output_text = gr.outputs.Textbox(label="Caption")
#examples = [[f"Image{i}.png" ] + ["Unconditional", ""] for i in range(1, 4)]
examples = [["Image1.png","Conditional", "Goal"],
           ["Image2.png","Unconditional", ""],
           ["Image3.png","Conditional", "Watch"]]

gr.Interface(fn=generate_caption, inputs=[input_image, radio_button, input_text], outputs=output_text, examples = examples, title="Image Captioning", description = "Model was taken from https://huggingface.co/Salesforce/blip-image-captioning-base.You can provide words as conditions to give a direction to your caption.You can refer to the examples below.").launch()