# HSI-Road Relabeled: Surface-Aware Road-Scene Segmentation

arXiv: https://arxiv.org/abs/2609.12151 · HTML: https://mlsanity.com/paper/2609.12151/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 7 of its 10 references that are in the map.
- First posted on arXiv 10 Sep 2026.

## Builds on (most cited first, 7 of 7)

- [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://mlsanity.com/paper/1505.04597/) (2015)
- [DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs](https://mlsanity.com/paper/1606.00915/) (2016)
- [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://mlsanity.com/paper/2105.15203/) (2021)
- [CBAM: Convolutional Block Attention Module](https://mlsanity.com/paper/1807.06521/) (2018)
- [Unified Perceptual Parsing for Scene Understanding](https://mlsanity.com/paper/1807.10221/) (2018)
- [Hyperspectral Imaging-Based Perception in Autonomous Driving Scenarios: Benchmarking Baseline Semantic Segmentation Models](https://mlsanity.com/paper/2410.22101/) (2024)
- [Hyperspectral Sensors and Autonomous Driving: Technologies, Limitations, and Opportunities](https://mlsanity.com/paper/2508.19905/) (2025)

## Abstract (arXiv)

The HSI-Road dataset provides paired RGB and 25-channel NIR (680-960nm) images with binary masks but no surface-level labels. This paper introduces a manually labeled six-class taxonomy: Background, Asphalt, Concrete, Dirt, Water, and Grass, and an RGB-to-NIR registration pipeline with corresponding annotations. Six semantic-segmentation models are evaluated under four input configurations: original-resolution RGB, registered low-resolution RGB (RGBreg), NIR, and channel-stacked RGBreg-NIR (RGBNstk). The comparison quantifies the effect of spatial-resolution reduction on RGB, along with evaluation of NIR and RGBNstk, with results reported using per-class and mean IoU and F1 scores. The original-resolution RGB achieves the highest overall performance but contains 12× more pixels and incurs a 15.5-20.6% latency penalty compared to the reduced-resolution inputs. At the common 192×384 resolution, RGBNstk outperforms NIR for all six models and RGBreg for five of six, with the most consistent gains for the Water class. These results highlight the importance of spatial resolution while showing that NIR provides complementary information to RGB.