示例

这个模型并不是最先进的模型，但仍然能够产生合理的图像标题结果。它主要是作为对 ? FlaxVisionEncoderDecoder 框架的概念验证而进行微调的。

这个模型可以按照以下方式使用：

在 PyTorch 中

import torch
import requests
from PIL import Image
from transformers import ViTFeatureExtractor, AutoTokenizer, VisionEncoderDecoderModel


loc = "ydshieh/vit-gpt2-coco-en"

feature_extractor = ViTFeatureExtractor.from_pretrained(loc)
tokenizer = AutoTokenizer.from_pretrained(loc)
model = VisionEncoderDecoderModel.from_pretrained(loc)
model.eval()


def predict(image):

    pixel_values = feature_extractor(images=image, return_tensors="pt").pixel_values

    with torch.no_grad():
        output_ids = model.generate(pixel_values, max_length=16, num_beams=4, return_dict_in_generate=True).sequences

    preds = tokenizer.batch_decode(output_ids, skip_special_tokens=True)
    preds = [pred.strip() for pred in preds]

    return preds


# We will verify our results on an image of cute cats
url = "http://images.cocodataset.org/val2017/000000039769.jpg"
with Image.open(requests.get(url, stream=True).raw) as image:
    preds = predict(image)

print(preds)
# should produce
# ['a cat laying on top of a couch next to another cat']

在 Flax 中

import jax
import requests
from PIL import Image
from transformers import ViTFeatureExtractor, AutoTokenizer, FlaxVisionEncoderDecoderModel


loc = "ydshieh/vit-gpt2-coco-en"

feature_extractor = ViTFeatureExtractor.from_pretrained(loc)
tokenizer = AutoTokenizer.from_pretrained(loc)
model = FlaxVisionEncoderDecoderModel.from_pretrained(loc)

gen_kwargs = {"max_length": 16, "num_beams": 4}


# This takes sometime when compiling the first time, but the subsequent inference will be much faster
@jax.jit
def generate(pixel_values):
    output_ids = model.generate(pixel_values, **gen_kwargs).sequences
    return output_ids
    
    
def predict(image):

    pixel_values = feature_extractor(images=image, return_tensors="np").pixel_values
    output_ids = generate(pixel_values)
    preds = tokenizer.batch_decode(output_ids, skip_special_tokens=True)
    preds = [pred.strip() for pred in preds]
    
    return preds
    
    
# We will verify our results on an image of cute cats
url = "http://images.cocodataset.org/val2017/000000039769.jpg"
with Image.open(requests.get(url, stream=True).raw) as image:
    preds = predict(image)
    
print(preds)
# should produce
# ['a cat laying on top of a couch next to another cat']