README.md

metadata

license: apache-2.0

OFA-tiny

Introduction

This is the tiny version of OFA pretrained model finetuned on vqaV2.

The directory includes 4 files, namely config.json which consists of model configuration, vocab.json and merge.txt for our OFA tokenizer, and lastly pytorch_model.bin which consists of model weights.

How to use

Download the models as shown below.

git clone https://github.com/sohananisetty/OFA_VQA.git
git clone https://huggingface.co/SohanAnisetty/ofa-vqa-tiny

After, refer the path to ofa-vqa-tiny to ckpt_dir, and prepare an image for the testing example below.

>>> from PIL import Image
>>> from torchvision import transforms
>>> from transformers import OFATokenizer, OFAModelForVQA

>>> mean, std = [0.5, 0.5, 0.5], [0.5, 0.5, 0.5]
>>> resolution = 256
>>> patch_resize_transform = transforms.Compose([
        lambda image: image.convert("RGB"),
        transforms.Resize((resolution, resolution), interpolation=Image.BICUBIC),
        transforms.ToTensor(), 
        transforms.Normalize(mean=mean, std=std)
    ])


>>> tokenizer = OFATokenizer.from_pretrained(ckpt_dir)

>>> txt = " what does the image describe?"
>>> inputs = tokenizer([txt], return_tensors="pt").input_ids
>>> img = Image.open(path_to_image)
>>> patch_img = patch_resize_transform(img).unsqueeze(0)


>>> model = OFAModel.from_pretrained(ckpt_dir, use_cache=False)
>>> gen = model.generate(inputs, patch_images=patch_img, num_beams=5, no_repeat_ngram_size=3) 

>>> print(tokenizer.batch_decode(gen, skip_special_tokens=True))