Papers › HS3-Bench: A Benchmark and Strong Baseline for Hyperspectral Semantic Segmentation in...

HS3-Bench: A Benchmark and Strong Baseline for Hyperspectral Semantic Segmentation in Driving Scenarios

17 Sep 2024arXiv:2409.11205archive 2025-07-28

Nick Theisen, Robin Bartsch, Dietrich Paulus, Peer Neubert

Semantic segmentation is an essential step for many vision applications in order to understand a scene and the objects within. Recent progress in hyperspectral imaging technology enables the application in driving scenarios and the hope is that the devices perceptive abilities provide an advantage over RGB-cameras. Even though some datasets exist, there is no standard benchmark available to systematically measure progress on this task and evaluate the benefit of hyperspectral data. In this paper, we work towards closing this gap by providing the HyperSpectral Semantic Segmentation benchmark (HS3-Bench). It combines annotated hyperspectral images from three driving scenario datasets and provides standardized metrics, implementations, and evaluation protocols. We use the benchmark to derive two strong baseline models that surpass the previous state-of-the-art performances with and without pre-training on the individual datasets. Further, our results indicate that the existing learning-based methods benefit more from leveraging additional RGB training data than from leveraging the additional hyperspectral channels. This poses important questions for future research on hyperspectral imaging for semantic segmentation in driving scenarios. Code to run the benchmark and the strong baseline approaches are available under https://github.com/nickstheisen/hyperseg.

PaperPDFCode

Code

nickstheisen/hyperseg officialmentioned in papermentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Autonomous DrivingHyperspectral Image SegmentationHyperspectral Semantic SegmentationSegmentationSemantic Segmentation

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Hyperspectral Semantic Segmentation HSI-Drive v2.0 RU-Net Accuracy 96.08 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HSI-Drive v2.0 RU-Net Average Accuracy 79.82 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HSI-Drive v2.0 RU-Net Avg. F1 82.34 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HSI-Drive v2.0 RU-Net Jaccard (Mean) 72.18 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HSI-Drive v2.0 U-Net Accuracy 94.95 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HSI-Drive v2.0 U-Net Average Accuracy 74.74 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HSI-Drive v2.0 U-Net Avg. F1 76.08 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HSI-Drive v2.0 U-Net Jaccard (Mean) 64.95 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HSI-Drive v2.0 DeepLabV3+ Accuracy 92.51 #3 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HSI-Drive v2.0 DeepLabV3+ Average Accuracy 65.58 #3 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HSI-Drive v2.0 DeepLabV3+ Avg. F1 67.86 #3 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HSI-Drive v2.0 DeepLabV3+ Jaccard (Mean) 56.63 #3 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS RU-Net Accuracy 86.72 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS RU-Net Average Accuracy 68.79 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS RU-Net Average Jaccard 58.64 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS RU-Net Avg. F1 69.19 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS U-Net Accuracy 85.36 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS U-Net Average Accuracy 68.15 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS U-Net Average Jaccard 57.39 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS U-Net Avg. F1 68.55 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS DeepLabV3+ Accuracy 84.10 #3 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS DeepLabV3+ Average Accuracy 63.01 #3 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS DeepLabV3+ Average Jaccard 53.22 #3 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation HyKo2-VIS DeepLabV3+ Avg. F1 64.90 #3 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City RU-Net Accuracy 87.63 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City RU-Net Average Accuracy 54.14 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City RU-Net Avg. F1 53.26 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City RU-Net Jaccard (Mean) 43.33 #1 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City DeepLabV3+ Accuracy 86.60 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City DeepLabV3+ Average Accuracy 53.15 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City DeepLabV3+ Avg. F1 51.83 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City DeepLabV3+ Jaccard (Mean) 40.79 #2 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City U-Net Accuracy 85.25 #3 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City U-Net Average Accuracy 48.62 #3 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City U-Net Avg. F1 48.18 #3 of 3 Archive leaderboard report
Hyperspectral Semantic Segmentation Hyperspectral City U-Net Jaccard (Mean) 37.73 #3 of 3 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections