许可证：Apache-2.0
任务类型：图像转文本
标签：

• 图像描述生成
• 图像转文本

世萌验证码识别模型，基于60000张图片训练集微调Vit-GPT2架构

Transformers调用方式

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

# 加载预训练模型
model = VisionEncoderDecoderModel.from_pretrained("AIris-Channel/vit-gpt2-verifycode-caption")
feature_extractor = ViTImageProcessor.from_pretrained("AIris-Channel/vit-gpt2-verifycode-caption")
tokenizer = AutoTokenizer.from_pretrained("AIris-Channel/vit-gpt2-verifycode-caption")

# 设备配置
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# 生成参数设置
max_length = 16
num_beams = 4
gen_kwargs = {"max_length": max_length, "num_beams": num_beams}

def predict_step(image_paths):
    images = []
    for image_path in image_paths:
        # 图像预处理
        i_image = Image.open(image_path)
        if i_image.mode != "RGB":
            i_image = i_image.convert(mode="RGB")
        images.append(i_image)

    # 特征提取与预测
    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)

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

# 示例调用
pred = predict_step(['ZZZTVESE.jpg'])
print(pred)  # 输出：zzztvese