66 lines
2.1 KiB
Markdown
66 lines
2.1 KiB
Markdown
|
|
# Documentation
|
|
- Class name: DSINE-NormalMapPreprocessor
|
|
- Category: ControlNet Preprocessors/Normal and Depth Estimators
|
|
- Output node: False
|
|
|
|
DSINE Normal Map预处理器节点旨在使用DSINE模型预处理图像以生成法线贴图。它通过调整视场角和迭代参数来优化模型的性能,从而从输入图像中生成详细的法线贴图。
|
|
|
|
# Input types
|
|
## Required
|
|
- image
|
|
- 图像输入是生成法线贴图的目标,作为模型应用处理并生成法线贴图的基础。
|
|
- Comfy dtype: IMAGE
|
|
- Python dtype: torch.Tensor
|
|
|
|
## Optional
|
|
- fov
|
|
- 视场角参数调整生成法线贴图的视角,影响结果图像中的深度感知和细节。
|
|
- Comfy dtype: FLOAT
|
|
- Python dtype: float
|
|
- iterations
|
|
- 迭代参数控制模型执行的优化步骤数量,影响生成的法线贴图的准确性和细节程度。
|
|
- Comfy dtype: INT
|
|
- Python dtype: int
|
|
- resolution
|
|
- 分辨率参数指定生成的法线贴图的输出分辨率,影响细节水平和清晰度。
|
|
- Comfy dtype: INT
|
|
- Python dtype: int
|
|
|
|
# Output types
|
|
- image
|
|
- 输出是一个法线贴图图像,它以颜色形式直观地表示表面法线,提供输入图像表面几何形状的详细信息。
|
|
- Comfy dtype: IMAGE
|
|
- Python dtype: Tuple[torch.Tensor]
|
|
|
|
|
|
## Usage tips
|
|
- Infra type: `GPU`
|
|
- Common nodes: unknown
|
|
|
|
|
|
## Source code
|
|
```python
|
|
class DSINE_Normal_Map_Preprocessor:
|
|
@classmethod
|
|
def INPUT_TYPES(s):
|
|
return create_node_input_types(
|
|
fov=("FLOAT", {"min": 0.0, "max": 365.0, "step": 0.05, "default": 60.0}),
|
|
iterations=("INT", {"min": 1, "max": 20, "step": 1, "default": 5})
|
|
)
|
|
|
|
RETURN_TYPES = ("IMAGE",)
|
|
FUNCTION = "execute"
|
|
|
|
CATEGORY = "ControlNet Preprocessors/Normal and Depth Estimators"
|
|
|
|
def execute(self, image, fov, iterations, resolution=512, **kwargs):
|
|
from controlnet_aux.dsine import DsineDetector
|
|
|
|
model = DsineDetector.from_pretrained().to(model_management.get_torch_device())
|
|
out = common_annotator_call(model, image, fov=fov, iterations=iterations, resolution=resolution)
|
|
del model
|
|
return (out,)
|
|
|
|
```
|