[Inference.Core] DensePose Estimator

Documentation

• Class name: Inference_Core_DensePosePreprocessor
• Category: ControlNet Preprocessors/Faces and Poses Estimators
• Output node: False

The DensePose Preprocessor node is designed to estimate human body poses and overlay them on images using DensePose models. It supports different models and color maps for visualization, providing a flexible approach to pose estimation and visualization.

Input types

Required

• image
  • The input image on which body poses will be estimated and visualized.
  • Comfy dtype: IMAGE
  • Python dtype: torch.Tensor

Optional

• model
  • Specifies the DensePose model to be used for pose estimation. The choice of model can affect the accuracy and performance of pose estimation.
  • Comfy dtype: COMBO[STRING]
  • Python dtype: str
• cmap
  • Determines the color map used for visualizing the pose estimation results. Different color maps can be used to enhance the visual distinction of pose estimations.
  • Comfy dtype: COMBO[STRING]
  • Python dtype: str
• resolution
  • The resolution to which the input image is resized before pose estimation. Affects the detail level of the pose estimation.
  • Comfy dtype: INT
  • Python dtype: int

Output types

• image
  • Comfy dtype: IMAGE
  • The output is an image with human body poses estimated and visualized on it.
  • Python dtype: torch.Tensor

Usage tips

• Infra type: GPU
• Common nodes: unknown

Source code

class DensePose_Preprocessor:
    @classmethod
    def INPUT_TYPES(s):
        return create_node_input_types(
            model=(["densepose_r50_fpn_dl.torchscript", "densepose_r101_fpn_dl.torchscript"], {"default": "densepose_r50_fpn_dl.torchscript"}),
            cmap=(["Viridis (MagicAnimate)", "Parula (CivitAI)"], {"default": "Viridis (MagicAnimate)"})
        )

    RETURN_TYPES = ("IMAGE",)
    FUNCTION = "execute"

    CATEGORY = "ControlNet Preprocessors/Faces and Poses Estimators"

    def execute(self, image, model, cmap, resolution=512):
        from controlnet_aux.densepose import DenseposeDetector
        model = DenseposeDetector \
                    .from_pretrained(filename=model) \
                    .to(model_management.get_torch_device())
        return (common_annotator_call(model, image, cmap="viridis" if "Viridis" in cmap else "parula", resolution=resolution), )