File size: 1,617 Bytes
759ce14 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 |
import torch
from PIL import Image
from transformers import (AutoTokenizer, VisionEncoderDecoderModel,
ViTFeatureExtractor)
import gradio as gr
if torch.cuda.is_available():
device = "cuda"
else:
device = "cpu"
encoder_checkpoint = "google/vit-base-patch16-224-in21k"
decoder_checkpoint = "distilgpt2"
model_checkpoint = "gagan3012/ViTGPT2_vizwiz"
feature_extractor = ViTFeatureExtractor.from_pretrained(encoder_checkpoint)
tokenizer = AutoTokenizer.from_pretrained(decoder_checkpoint)
model = VisionEncoderDecoderModel.from_pretrained(model_checkpoint).to(device)
def predict(image):
clean_text = lambda x: x.replace("<|endoftext|>", "").split("\n")[0]
sample = feature_extractor(image, return_tensors="pt").pixel_values.to(device)
caption_ids = model.generate(sample, max_length=50)[0]
caption_text = clean_text(tokenizer.decode(caption_ids))
return caption_text
inputs = [
gr.inputs.Image(type="pil", label="Original Image")
]
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 images"
article = " <a href='https://huggingface.co/gagan3012/ViTGPT2_vizwiz'>Model Repo on Hugging Face Model Hub</a>"
examples = [
["people-walking-street-pedestrian-crossing-traffic-light-city.jpeg"],
["elonmusk.jpeg"]
]
gr.Interface(
predict,
inputs,
outputs,
title=title,
description=description,
article=article,
examples=examples,
theme="huggingface",
).launch(debug=True, enable_queue=True)
|