app.py

import os
os.environ['HF_HUB_DISABLE_SYMLINKS_WARNING'] = '1'

import torch
from torchvision import models, transforms
from PIL import Image
from transformers import GPT2LMHeadModel, GPT2Tokenizer
import streamlit as st

# Load the pre-trained image feature extraction model
resnet = models.resnet50(pretrained=True)
resnet.eval()

# Load the pre-trained language model
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
model = GPT2LMHeadModel.from_pretrained("gpt2")
model.eval()

# Preprocess the image
def preprocess_image(image_path):
    image = Image.open(image_path)
    preprocess = transforms.Compose([
        transforms.Resize(256),
        transforms.CenterCrop(224),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
    ])
    input_tensor = preprocess(image)
    input_batch = input_tensor.unsqueeze(0)
    return input_batch

# Extract image features
def extract_image_features(image_path):
    input_batch = preprocess_image(image_path)
    with torch.no_grad():
        output = resnet(input_batch)
    image_features = output.squeeze(0)
    return image_features

# Generate caption
def generate_caption(image_features):
    caption = tokenizer.decode(model.generate(input_ids=model.config.pad_token_id,
                                               max_length=50,
                                               eos_token_id=model.config.eos_token_id,
                                               no_repeat_ngram_size=2,
                                               num_return_sequences=1,
                                               attention_mask=None,
                                               encoder_outputs=None,
                                               decoder_start_token_id=None,
                                               use_cache=None,
                                               labels=None,
                                               output_attentions=None,
                                               output_hidden_states=None,
                                               output_scores=None,
                                               return_dict=None,
                                               input_embeds=image_features.unsqueeze(0)))
    return caption

# Streamlit app
st.title("Image Captioning with GPT-2")
uploaded_file = st.file_uploader("Choose an image...", type="jpg")

if uploaded_file is not None:
    # Display the uploaded image
    image = Image.open(uploaded_file)
    st.image(image, caption="Uploaded Image", use_column_width=True)

    # Generate caption when the image is uploaded
    image_features = extract_image_features(uploaded_file)
    caption = generate_caption(image_features)
    st.write("Generated Caption:", caption)