Yet another custom node to detect human parts

Detect human parts using the DeepLabV3+ ResNet50 model from Keras-io. You can extract hair, arms, legs, and other parts with ease and with small memory usage.

This node aims to detect human parts using the model created by Keras-io. Their "Space" was impressive, and I wanted to use the model.

Unfortunately, the model uses an old Keras version, and there were no PyTorch implementation.

So I decided to convert the model to ONNX format and to create my HugginFace repository to share the model with the community.

Fortunately, Keras provides the model with a CC1.0 license, thank you guys to allow us to use it without any restriction.

Example

You can drag and drop the following image to try:


DeepLabV3+ ResNet50 for Human

Actually, all the model I found was not trained to detect human parts, but to detect some objects or urban elements. The Keras model is the only one I found that works!

Installation

I strongly recommend to use ComfyUI-Manager to install the node. It will install the dependencies and the model.

If you're using the command line, you can install the node with:

cd /path/to/your/ComfyUI/custom_nodes
git clone ...
cd HumanParts
pip install -r requirements.txt
# or
python -m pip install -r requirements.txt

Then, restart ComfyUI and you may find the "HumanParts" node.