标签:

• 图像转文本
• 图像描述生成 语言:
• 俄语 评估指标:
• 双语评估替补分数（BLEU） 库名称: transformers

首个俄语图像描述生成模型 vit-rugpt2-image-captioning

这是一个基于COCO2014数据集翻译版（英俄）训练的图片描述生成模型。

模型详情

模型编码器部分采用google/vit-base-patch16-224-in21k初始化，解码器部分采用sberbank-ai/rugpt3large_based_on_gpt2初始化。

测试数据指标

• BLEU分数: 8.672
• BLEU-1精确率: 30.567
• BLEU-2精确率: 7.895
• BLEU-3精确率: 3.261

示例运行代码

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

model = VisionEncoderDecoderModel.from_pretrained("vit-rugpt2-image-captioning")
feature_extractor = ViTFeatureExtractor.from_pretrained("vit-rugpt2-image-captioning")
tokenizer = AutoTokenizer.from_pretrained("vit-rugpt2-image-captioning")

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

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

def predict_caption(image_paths):
  images = []
  for image_path in image_paths:
    i_image = Image.open(image_path)
    if i_image.mode != "RGB":
      i_image = i_image.convert(mode="RGB")

    images.append(i_image)

  pixel_values = feature_extractor(images=images, return_tensors="pt").pixel_values
  pixel_values = pixel_values.to(device)

  output_ids = model.generate(pixel_values, **gen_kwargs)

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

predict_caption(['train2014/COCO_train2014_000000295442.jpg']) # ['飞机停在机场跑道上。']

使用transformers流水线的示例代码

from transformers import pipeline

image_to_text = pipeline("image-to-text", model="vit-rugpt2-image-captioning")

image_to_text("train2014/COCO_train2014_000000296754.jpg") # [{'generated_text': '一个人撑着伞走在街上。'}]

获取帮助联系方式

• https://huggingface.co/tuman
• https://github.com/tumanov-a
• https://t.me/tumanov_av