Vit画patch块代码实现

做出Vit的patch块，用于Visio画图，如下所示：

\

代码实现：

from PIL import Image

# 读取原始图像
img_path = 'input/0001_c016_00012925_0_orgin.jpg'
img = Image.open(img_path)

# 将图像调整到150x150像素大小（如果原始图像不是这样）
img_resized = img.resize((240, 240))

# 每个patch的大小和空隙大小
patch_size = 30
gap_size = 5
patch_num = 8

# 计算画布的尺寸
canvas_width = (patch_size + gap_size) * patch_num - gap_size
canvas_height = (patch_size + gap_size) * patch_num - gap_size

# 创建拼接画布，白色背景
stitched_image = Image.new('RGB', (canvas_width, canvas_height), (255, 255, 255))

# 截取并拼接带空隙的patch
for i in range(patch_num):
    for j in range(patch_num):
        # 截取30x30的区域
        patch = img_resized.crop((j * patch_size, i * patch_size, (j + 1) * patch_size, (i + 1) * patch_size))
        # 计算画布上的粘贴位置
        # patch.save(f'output/patch_mask_{i * 5 + j + 1}.jpg')
        pos_x = j * (patch_size + gap_size)
        pos_y = i * (patch_size + gap_size)
        # 粘贴patch到画布上，不需要mask
        stitched_image.paste(patch, (pos_x, pos_y))

# 保存拼接后的图像
stitched_image.save(f'output/patch_{img_path.split("/")[-1].split(".")[0]}.jpg')
stitched_image.show()

上述代码，根据自己设定的patch块数量，微调resize，patch_size和缝隙大小即可

上述代码执行后效果为：


pathch_num=5,图像大小设置为（150，150）后：