ApacheOne / ComfyUI-human-parser_models_ATR_LIP_Pascal

As always : More safe for everyone to share around and keep updated if any major changes then the google drive.

Copy from github fork:

• human-parser-comfyui-node-in-pure-python
• This custom node doesn't require a runtime build for InPlaceABNSync.
• original Github repo
• Needs C++ build tools for windows

Potential fix

• for comfyUI node: "cant find schp"(model files in folder)
• replace init .py from the pure python fork with the one in this repo here initfix
• have not tested C++ version might have same fix

cloned README.md

Cozy Human Parser

Fast, VRAM-light ComfyUI nodes to generate masks for specific body parts and clothes or fashion items. Runs on CPU and CUDA. Made with 💚 by the CozyMantis squad.

Original	ATR	LIP	Pascal

Installation

• Clone this repository into your custom_nodes directory, then run pip install -r requirements.txt to install the required dependencies.
• Copy the following models to the models/schp directory, depending on which parser you would like to use:
  • Model based on the LIP dataset: Google Drive
  • Model based on the ATR dataset: Google Drive
  • Model based on the Pascal dataset: Google Drive

Check below for Windows troubleshooting .

Examples

LIP Parser

• LIP is the largest single person human parsing dataset with 50000+ images. This dataset focuses on complicated real scenarios.
• mIoU on LIP validation: 59.36 %
• The LIP parser can detect the following categories:

['Background', 'Hat', 'Hair', 'Glove', 'Sunglasses' 'Upper-clothes', 'Dress', 'Coat', 'Socks', 'Pants', 'Jumpsuits', 'Scarf', 'Skirt', 'Face', 'Left-arm', 'Right-arm', 'Left-leg', 'Right-leg', 'Left-shoe', 'Right-shoe']

ATR Parser

• ATR is a large single person human parsing dataset with 17000+ images. This dataset focuses on fashion AI.
• mIoU on ATR test: 82.29%
• The ATR parser can detect the following categories:

['Background', 'Hat', 'Hair', 'Sunglasses', 'Upper-clothes', 'Skirt', 'Pants', 'Dress', 'Belt', 'Left-shoe', 'Right-shoe', 'Face', 'Left-leg', 'Right-leg', 'Left-arm', 'Right-arm', 'Bag', 'Scarf']

Pascal Parser

• Pascal Person Part is a tiny single person human parsing dataset with 3000+ images. This dataset focuses on body parts segmentation.
• mIoU on Pascal-Person-Part validation: 71.46 %
• The Pascal parser can detect the following categories:

['Background', 'Head', 'Torso', 'Upper Arms', 'Lower Arms', 'Upper Legs', 'Lower Legs']

Windows Troubleshooting

• Ninja is required to load C++ extensions

Windows can't find the "ninja.exe" file. The file is probably getting downloaded/installed to something like X:\path\to\comfy\python_embeded\lib\site-packages\ninja\data\bin , but it's not properly getting added to the system path, so the OS can't invoke it.

The solution is to:

• locate the "ninja.exe" file;

• add the full path to ninja.exe into the system PATH:

  • see https://www.mathworks.com/matlabcentral/answers/94933-how-do-i-edit-my-system-path-in-windows
  • remember, you need to enter the path to the folder containing the ninja.exe binary)
  • see this issue for more details

• NK1104: cannot open file 'python311.lib' (or similar)

Windows can't locate the python311.lib library. You need to search for it on your system, then add the library's parent directory to the "LIB" environment variable.

• see this SO article

• also see this issue

• Command '['where', 'cl']' returned non-zero exit status 1

Windows can't locate "cl.exe" which is the compiler/linker tool: https://learn.microsoft.com/en-us/cpp/build/reference/compiler-options?view=msvc-170

You can start this tool only from a Visual Studio developer command prompt. You cannot start it from a system command prompt or from File Explorer. For more information, see Use the MSVC toolset from the command line.

First, make sure you've installed all of the things highlighted below:

Then, it looks like you'll need to start ComfyUI from the developer command prompt instead of the regular cmd. Here's docs on how to launch the dev command prompt: https://learn.microsoft.com/en-us/visualstudio/ide/reference/command-prompt-powershell?view=vs-2022

You'll want to run something similar to:

cd X:\path\to\comfy
python main.py

• error: first parameter of allocation function must be of type "size_t"

Make sure you're running the "x64 Native Tools Command Prompt" instead of the x86 one. Type "x64" in the start menu to locate it.

Acknowledgements

-All the repo owners and maintainers Based on the excellent paper "Self-Correction for Human Parsing" by Li, Peike and Xu, Yunqiu and Wei, Yunchao and Yang, Yi, and their original code that we've updated to also run on CPUs.

@article{li2020self,
  title={Self-Correction for Human Parsing}, 
  author={Li, Peike and Xu, Yunqiu and Wei, Yunchao and Yang, Yi},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, 
  year={2020},
  doi={10.1109/TPAMI.2020.3048039}}