112 lines
3.6 KiB
Markdown
112 lines
3.6 KiB
Markdown
|
|
# Documentation
|
|
- Class name: ImageNoiseBinomial
|
|
- Category: image/noise
|
|
- Output node: False
|
|
- Repo Ref: https://github.com/comfyanonymous/ComfyUI
|
|
|
|
ImageNoiseBinomial节点对图像应用二项分布噪声，允许调整单色性、反转和特定通道。它操作图像数据以引入随机变化，模拟基于二项分布噪声对视觉内容的影响。这种方法可以用于创造各种视觉效果，从细微的纹理增强到更显著的图像扰动。
|
|
|
|
# Input types
|
|
## Required
|
|
- images
|
|
- 'images'参数代表将要应用二项分布噪声的输入图像。它是定义将进行噪声添加的基础数据的关键。
|
|
- Comfy dtype: IMAGE
|
|
- Python dtype: torch.Tensor
|
|
- n
|
|
- 'n'参数指定二项分布噪声生成过程中的试验次数，影响噪声在图像中的强度和分布。
|
|
- Comfy dtype: INT
|
|
- Python dtype: int
|
|
- p
|
|
- 'p'参数定义二项分布中每次试验的成功概率，影响图像中噪声的整体外观。
|
|
- Comfy dtype: FLOAT
|
|
- Python dtype: float
|
|
- monochromatic
|
|
- 'monochromatic'参数决定噪声是否应均匀地应用于所有通道或每个通道都有变化，影响噪声的视觉一致性。
|
|
- Comfy dtype: COMBO[STRING]
|
|
- Python dtype: bool
|
|
- invert
|
|
- 'invert'参数决定是否应该反转噪声效果，提供了一种从图像中减去噪声而不是添加噪声的方法，改变视觉结果。
|
|
- Comfy dtype: COMBO[STRING]
|
|
- Python dtype: bool
|
|
- channels
|
|
- 'channels'参数指定应该将噪声应用于图像的哪些颜色通道，允许有针对性的噪声应用和对视觉效果的控制。
|
|
- Comfy dtype: COMBO[STRING]
|
|
- Python dtype: str
|
|
|
|
# Output types
|
|
- image
|
|
- 输出是应用了二项分布噪声的图像或一组图像，反映了指定参数对原始输入的影响。
|
|
- Comfy dtype: IMAGE
|
|
- Python dtype: torch.Tensor
|
|
|
|
|
|
## Usage tips
|
|
- Infra type: `GPU`
|
|
- Common nodes: unknown
|
|
|
|
|
|
## Source code
|
|
```python
|
|
class ImageNoiseBinomial:
|
|
def __init__(self):
|
|
pass
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"images": ("IMAGE",),
|
|
"n": ("INT", {
|
|
"default": 128,
|
|
"min": 1,
|
|
"max": 255,
|
|
}),
|
|
"p": ("FLOAT", {
|
|
"default": 0.5,
|
|
"max": 1.0,
|
|
"step": 0.01
|
|
}),
|
|
"monochromatic": (["false", "true"],),
|
|
"invert": (["false", "true"],),
|
|
"channels": (["rgb", "rgba", "rg", "rb", "ra", "gb", "ga", "ba", "r", "g", "b", "a"],),
|
|
},
|
|
}
|
|
|
|
RETURN_TYPES = ("IMAGE",)
|
|
FUNCTION = "node"
|
|
CATEGORY = "image/noise"
|
|
|
|
def noise(self, images, n, p, monochromatic, invert):
|
|
if monochromatic and images.shape[3] > 1:
|
|
noise = np.random.binomial(n, p, images.shape[:3])
|
|
else:
|
|
noise = np.random.binomial(n, p, images.shape)
|
|
|
|
noise = noise.astype(images.dtype)
|
|
noise /= 255
|
|
|
|
if monochromatic and images.shape[3] > 1:
|
|
noise = noise[..., np.newaxis].repeat(images.shape[3], -1)
|
|
|
|
if invert:
|
|
noise = images - noise
|
|
else:
|
|
noise = images + noise
|
|
|
|
noise = np.clip(noise, 0.0, 1.0)
|
|
|
|
return noise
|
|
|
|
def node(self, images, n, p, monochromatic, invert, channels):
|
|
tensor = images.clone().detach()
|
|
|
|
monochromatic = True if monochromatic == "true" else False
|
|
invert = True if invert == "true" else False
|
|
|
|
return (channels_layer(tensor, channels, lambda x: self.noise(
|
|
x, n, p, monochromatic, invert
|
|
)),)
|
|
|
|
```
|