Nicr Scene Analysis: Preparation and Loading Tools for Scene Datasets

TUI-NICR released this repository in 2025 to provide standardized preparation and loading utilities for RGB-D scene analysis datasets. It facilitates the integration of multiple segmentation types and orientation estimation data specifically for mobile robotics research.

Use Cases

• Training panoptic segmentation models using RGB-D image pairs and instance masks.
• Developing orientation-aware object detection systems for mobile robotics using orientation-estimation labels.
• Benchmarking semantic segmentation performance across different scene analysis datasets using the unified PyTorch data loaders.

Strengths

• Supports three distinct segmentation types: panoptic, semantic, and instance.
• Includes specialized labels for orientation-estimation tasks.
• Apache-2.0 license allows for broad redistribution and commercial use.

Limitations

• Requires external data sourcing as the repository provides preparation code rather than raw data files.
• Specific record counts and class distributions depend on the underlying dataset being processed.

Provenance

Source

TUI-NICR (Technical University of Ilmenau, Neuroinformatics and Cognitive Robotics Lab)

Collection Method

Software-based preparation and standardization of existing scene analysis benchmarks.

Freshness

Updated as of October 2025.