app.py

# Visual question answering from
# https://learn.deeplearning.ai/courses/open-source-models-hugging-face/lesson/13/multimodal-visual-question-answering
#

from transformers import BlipForQuestionAnswering
model = BlipForQuestionAnswering.from_pretrained("Salesforce/blip-vqa-base")

from transformers import AutoProcessor
processor = AutoProcessor.from_pretrained("Salesforce/blip-vqa-base")

from PIL import Image

import gradio as gr

def answering(image, question):
    inputs = processor(image, question, return_tensors="pt")
    out = model.generate(**inputs)
    output = processor.decode(out[0], skip_special_tokens=True)
    return output

gr.close_all()

app = gr.Interface(fn=answering,
                   inputs=[gr.Image(label="Picture here", type="pil"),
                           gr.Textbox(label="Question about picture here")],
                           outputs=[gr.Textbox(label="Answer"),],
                           title="Harza's application for answering questions about picture'",
                           description="Harza's miracle application that can answer questions about given picuture!'",
                           allow_flagging="never")
app.launch()
gr.close_all()