TMM-SPORTS: Simultaneous Panoptic Odometry, Rendering, Tracking and Segmentation for Urban Scenes Understanding

October 28, 2025 · View on GitHub

We proposes a novel framework, named SPORTS, for holistic scene understanding via tightly integrating Video Panoptic Segmentation (VPS), Visual Odometry (VO), and Scene Rendering (SR) tasks into an iterative and unified perspective.

Note

Our code is mainly divided into three parts: VPS, VO, and SR.

  1. The VPS module includes both the baseline model and models integrated with various channel attention mechanisms.
  2. The VO module consists of localization, point cloud map reconstruction, and the code for matching segmentation results.
  3. The SR module contains point-based rendering methods.

The released code constitutes the core components.

"Please refer to the code in VPS/knet/att for the core geometric information fusion module. Please refer to PVO、Video-K-Net、 READ for environment configuration and dataset processing."

Acknowledgement

Some code snippets are borrowed from PVO、Video-K-Net、 READ、EMA. Great thanks to them!