OFA大字幕

介绍

这是OFA模型的大版本，用于图像字幕。OFA是一个统一的多模态预训练模型，将模态（即交叉模态、视觉、语言）和任务（例如图像生成、视觉定位、图像字幕、图像分类、文本生成等）统一到一个简单的序列到序列学习框架中。

目录包括4个文件，分别是config.json（包含模型配置）、vocab.json和merge.txt（用于OFA分词器），以及pytorch_model.bin（包含模型权重）。由于我们已经解决了Fairseq和transformers之间的不匹配问题，所以不必担心。

如何使用

要在transformers中使用它，请参阅 https://github.com/OFA-Sys/OFA/tree/feature/add_transformers 。安装transformers并按下面展示的方式下载模型。

git clone --single-branch --branch feature/add_transformers https://github.com/OFA-Sys/OFA.git
pip install OFA/transformers/
git clone https://huggingface.co/OFA-Sys/OFA-large-caption

然后，将OFA大模型的路径引用到ckpt_dir，并准备一个图像用于下面的测试示例。还要确保您的环境中安装了Pillow和torchvision。

>>> from PIL import Image
>>> from torchvision import transforms
>>> from transformers import OFATokenizer, OFAModel
>>> from generate import sequence_generator

>>> mean, std = [0.5, 0.5, 0.5], [0.5, 0.5, 0.5]
>>> resolution = 480
>>> 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)


# using the generator of fairseq version
>>> model = OFAModel.from_pretrained(ckpt_dir, use_cache=True)
>>> generator = sequence_generator.SequenceGenerator(
                    tokenizer=tokenizer,
                    beam_size=5,
                    max_len_b=16, 
                    min_len=0,
                    no_repeat_ngram_size=3,
                )
>>> data = {}
>>> data["net_input"] = {"input_ids": inputs, 'patch_images': patch_img, 'patch_masks':torch.tensor([True])}
>>> gen_output = generator.generate([model], data)
>>> gen = [gen_output[i][0]["tokens"] for i in range(len(gen_output))]

# using the generator of huggingface version
>>> 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))