README.md

metadata

metrics:
  - bleu
  - rouge
tags:
  - image-to-text
  - image-captioning
  - vision-transformer
  - ViT-B/16
language:
  - id
  - en

Sample running code

from transformers import VisionEncoderDecoderModel, ViTImageProcessor, GPT2Tokenizer
import torch
from PIL import Image
model = VisionEncoderDecoderModel.from_pretrained("evlinzxxx/best_model_ViTB16_GPT2")
feature_extractor = ViTImageProcessor.from_pretrained("evlinzxxx/best_model_ViTB16_GPT2")
tokenizer = GPT2Tokenizer.from_pretrained("evlinzxxx/best_model_ViTB16_GPT2")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
def show_image_and_captions(url):
  # get the image and display it
  display(load_image(url))
  # get the captions on various models
  our_caption = get_caption(model, image_processor, tokenizer, url)
  # print the captions
  print(f"Our caption: {our_caption}")
show_image_and_captions("/content/drive/MyDrive/try/test_400/gl_16.jpg") # ['navigate around the obstacle ahead adjusting your route to bypass the parked car.']