PST-PCQA: Low-Complexity Patch-based No-Reference Point Cloud Quality Metric exploiting Weighted Structure and Texture Features (TBC 2025)

Official repository to the article "Low-Complexity Patch-based No-Reference Point Cloud Quality Metric exploiting Weighted Structure and Texture Features" accepted for publication in IEEE Transactions on Broadcasting.

Instructions

The structure of the project should be like the following:

• 📁 best_models/WPC
  • 🧠 PST_PCQAModule_K_16.ckpt (model trained on WPC with 16 patches)
  • 🧠 PST_PCQAModule_K_8.ckpt (model trained on WPC with 8 patches)
• 📁 data (you have to download the datasets)
  • 📁 WPC
    • 📁 distorted
    • 🔢 mos.xls
  • 📁 SIAT-PCQD
    • 📁 distorted
    • 🔢 DMOS.xlsx
  • 📁 SJTU-PCQA
    • 📁 distorted
    • 🔢 MOS.xlsx
• 📄 environment.yml (Conda environment to import)
• 📄 model.py (proposed approach)
• 📄 patch_extraction.py (To extract patches before training)
• 📄 util.py
• 📓 training_val_test_{WPC,SJTU,SIAT}.py (code for training, validating, and testing PST-PCQA on WPC, SIAT, and SJTU datasets)

First, run patch_extraction.py to create, using WPC dataset as an example, 📁 data/WPC/distorted_npz_K folder containing the K patches for each point cloud. Then, it is possible to run the trainin_val_test_WPC.py script, targeting the created folder containing the patches.

Inference

The repository contains 📄 example_inference.ipynb notebook to provide a minimal example on how to extract patches and perform inference on the fly. In addition, a visualization of the patch scores is also available (Note: Visualization is computationally expensive). The model takes approximately 50 ms to perform inference on a single NVIDIA RTX 4070.

Authors

Michael Neri*, Federica Battisti°

*Faculty of Information Technology and Communication Sciences, Tampere University, Tampere, Finland

°Department of Information Engineering, University of Padova, Padua, Italy

Reference

If you use part of this code, please cite the following article

@ARTICLE{Neri_TBC_2025,
   author={Neri, Michael and Battisti, Federica},
   journal={IEEE Transactions on Broadcasting}, 
   title={Low-Complexity Patch-Based No-Reference Point Cloud Quality Metric Exploiting Weighted Structure and Texture Features}, 
   year={2025},
   volume={71},
   number={2},
   pages={631-640},
   doi={10.1109/TBC.2025.3553305}
  }