app.py

# Image captioning with ViT+GPT2
from PIL import Image
from transformers import VisionEncoderDecoderModel, ViTFeatureExtractor, PreTrainedTokenizerFast
import requests
model = VisionEncoderDecoderModel.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
vit_feature_extactor = ViTFeatureExtractor.from_pretrained("google/vit-base-patch16-224-in21k")
tokenizer = PreTrainedTokenizerFast.from_pretrained("distilgpt2")


#url = 'https://d2gp644kobdlm6.cloudfront.net/wp-content/uploads/2016/06/bigstock-Shocked-and-surprised-boy-on-t-113798588-300x212.jpg'
# with Image.open(requests.get(url, stream=True).raw) as img:
#     pixel_values = vit_feature_extactor(images=img, return_tensors="pt").pixel_values
# encoder_outputs = model.generate(pixel_values.to('cpu'),num_beams = 5)
# generated_senetences = tokenizer.batch_decode(encoder_outputs, skip_special_tokens=True,)
# generated_senetences
# generated_senetences[0].split(".")[0]


def vit2distilgpt2(img):
    pixel_values = vit_feature_extactor(images=img, return_tensors="pt").pixel_values
    encoder_outputs = generated_ids = model.generate(pixel_values.to('cpu'),num_beams=5)
    generated_senetences = tokenizer.batch_decode(encoder_outputs, skip_special_tokens=True)

    return(generated_senetences[0].split('.')[0])

import gradio as gr
inputs = [
    gr.inputs.Image(type="pil",label="Original Images")
]

outputs = [
    gr.outputs.Textbox(label = "Caption")
]

title = "Image Captioning using ViT + GPT2"
description = "ViT and GPT2 are used to generate Image Caption for the uploaded image.COCO DataSet is used for Training"
examples = [
    ["Image1.png"],
    ["Image2.png"],
    ["Image3.png"]
]


gr.Interface(
    vit2distilgpt2,
    inputs,
    outputs,
    title=title,
    description=description,
    examples=examples,
    theme="huggingface",
).launch(debug=True, enable_queue=True)