app.py

import transformers, accelerate
import requests


print(accelerate.__version__)
print(transformers.__version__)

# Image Captioning

from transformers import AutoProcessor
from transformers import AutoModelForCausalLM
import torch
import streamlit as st

device = "cuda" if torch.cuda.is_available() else "cpu" # Set device to GPU if its available

checkpoint = "microsoft/git-base"
processor = AutoProcessor.from_pretrained(checkpoint) # We would load a tokenizer for language. Here we load a processor to process images

model = AutoModelForCausalLM.from_pretrained(checkpoint).to(device)


# Text Search
st.title("Flower Type Demo")
st.subheader("Upload an image and See how Chinese qisper works")

upload_file = st.file_uploader('Upload an Image')

from PIL import Image


if upload_file:
  test_sample = Image.open(upload_file)

  inputs = processor(images=image, return_tensors="pt").to(device)
  pixel_values = inputs.pixel_values.to(device)

  generated_ids = model.generate(pixel_values=pixel_values, max_length=50)
  generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]


  for i in range(10):
    st.write('New Caption is :')
    st.write(generated_caption)
    image = pipe(generated_caption).images[0]
    
    display(image)

    print("Model Loading + Inference time = " + str(time.time() - t1) + " seconds")

    st.write("Showing the Image")
    st.image (image, caption=name, width=None, use_column_width=None, clamp=False, channels='RGB', output_format='auto')

    inputs = processor(images=image, return_tensors="pt").to(device)
    pixel_values = inputs.pixel_values.to(device)
    generated_ids = model.generate(pixel_values=pixel_values, max_length=50)
    generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]