A Unified Benchmark for Aerial-Enhanced Autonomous Driving: The KITScenes Aerial Dataset

Daniel Lengerer; Richard Schwarzkopf; Jonas Merkert; Fabian Immel; Rupert Polley; Frank Bieder; Jan-Hendrik Pauls; J. Marius Zöllner

cs.CVOct 29, 2026

A Unified Benchmark for Aerial-Enhanced Autonomous Driving: The KITScenes Aerial Dataset

Authors: Daniel Lengerer, Richard Schwarzkopf, Jonas Merkert, Fabian Immel, Rupert Polley, Frank Bieder, Jan-Hendrik Pauls, J. Marius Zöllner,

Organizations: Technical University of Applied Sciences Augsburg · FZI Research Center for Information Technology · Karlsruhe Institute of Technology (KIT) · Parallel Domain ·

Abstract

Aerial imagery provides complementary spatial context for autonomous driving by exposing road structure beyond the immediate perception range of onboard sensors. Yet map construction from aerial imagery, aerial-assisted online HD mapping, and cross-view localization are typically studied on separate data and benchmarks. We introduce KITScenes Aerial, an extension of the KITScenes Multimodal dataset that augments synchronized onboard sensor data, vehicle poses, and rich Lanelet2 map annotations with spatially associated aerial imagery and higher-resolution neural representations, providing a common basis for a variety of research tasks. Our mapping evaluation focuses on all 53 KITScenes map classes across different model designs, spatial ranges, and aerial resolutions, while retaining compatibility with the widely used three-class setting. Representative baselines show that aerial context is particularly beneficial for fine-grained and extended-range mapping, more than doubling 53-class mAP at the larger mapping extent. Higher-resolution aerial imagery yields a further 4.3 mAP (+24 %) improvement, with gains extending to challenging map elements such as traffic lights. We additionally evaluate cross-view localization under geographically disjoint conditions, where an adapted fine-grained feature-matching baseline achieves a mean translation error of 5.13 m. KITScenes Aerial thus enables unified study of aerial and ground-level information across complementary mapping and localization tasks. Project page: https://kitscenes.com/aerial

Figures & tables

Figure 1: KITScenes Aerial: We extend KITScenes Multimodal with aligned aerial imagery with $20\text{\,}\mathrm{cm}\text{/}\mathrm{px}$ ground surface resolution (overlayed by the KITScenes HD map) and neural feature embeddings (bottom right, visualized as 3-channel PCA) for municipal high-resolution imagery with $8\text{\,}\mathrm{cm}\text{/}\mathrm{px}$ ground surface resolution (top right). Extending KITScenes Multimodal with these different aerial imagery sources enables multiple research directions currently underrepresented in autonomous driving datasets as shown in Tab. 1 .

Panel of Table 1 — Table 1: Overview of representative datasets and the research settings supported by their released data. Support is graded rather than binary: the released data supports the setting at full fidelity, support is limited, $\nsim$ all required data components are present but the aerial-to-ground alignment error exceeds the requirement of the setting, and – the setting is not supported. Pos.: georeferenced positions or poses. KITScenes Aerial is the only dataset that supports all four research settings at full fidelity.

Map Content

Structure

Aerial

Dataset

Map Representation

Cls

Topo

Reg

ROI

Feat

LaneExtraction [ 11 ]

Routable lane graph

–

✓

–

✓

–

tile

–

UrbanLaneGraph [ 2 ]

Directed centerline graph

–

✓

–

✓

–

tile

–

OpenSatMap [ 44 ]

Attributed vector primitives

✓

–

(✓)

–

tile

–

leveLX family 1

Lanelet2 / OpenDRIVE

n/r

✓

site

✓

SatForHDMap [ 8 ]

nuScenes vector map

✓

( ✓ ) 2

✓

( ✓ ) 2

–

✓

( ✓ ) 2

–

fix

–

Table 2: Map annotations and aerial provision of the datasets that combine map supervision with aerial imagery. Map content is reported over the six semantic categories used throughout our experiments: Lane Markings (LM), Lane Centerlines (LC), Road Infrastructure (RI), Traffic Lights (TL), Traffic Signs (TS) and Road Markings (RM). ✓ the category is part of the released map and of the mapping benchmark;

(\checkmark)

present in the underlying map but outside the released benchmark; – not annotated; ? not documented. Cls: number of benchmark label classes. Topo: lane connectivity. Reg: regulatory relations between traffic-control elements and lanes. ROI: form of the aerial region of interest ( tile freely selectable, fix fixed ego crop, site static per-site view). Feat: precomputed aerial features released. KITScenes Aerial is the only dataset whose benchmark covers all six categories, with topology and regulatory relations, on freely selectable aerial imagery.

Panel of Figure 2 — Figure 2: Qualitative ground–aerial correspondence in KITScenes Aerial. (a) LiDAR reflectivity basemap and (b) corresponding aerial overlay; (c,d) additional scenes. All ego-centered crops are extracted directly from the associated vehicle poses using identical spatial extents.

Panel of Figure 3 — Figure 3: Qualitative comparison of online HD map construction on KITScenes Aerial. Camera views provide the corresponding ground-level context. Predictions from the onboard-only StreamMapNet baseline are compared with the aerial-fused AerialFusionMapNet.

Lane Markings (LM)	Lane Centerlines (LC)	Road Infrastructure (RI)
road_border	centerline	fence
dashed	bike_centerline	building
…		…
Traffic Lights (TL)	Traffic Signs (TS)	Road Markings (RM)
tl_car	ts_stop	bike_symbol
tl_bike	ts_yield	bus_symbol

Table 3: For reporting, we aggregate the 53 KITScenes map-element classes into six categories, listed with representative entries.

Method	Training	Inference	AP LM	AP LC	AP RI	AP TL	AP TS	AP RM	mAP 53	$\mathrm{AP}_{\mathrm{Topo}}$
StreamMapNet [ 43 ]	C	C	$14.0$	$23.2$	$16.7$	$17.5$	$4.3$	$14.6$	$11.1$	$48.2$
StreamMapNet-CVS [ 20 ]	C + AID	C	$16.2$	$23.3$	$16.8$	$12.9$	$2.5$	$19.5$	$11.3$	$48.7$
AerialFusionMapNet [ 19 ]	C + AID	C + AID	$27.5$	$29.2$	$22.1$	$18.4$	$3.9$	$34.8$	$18.0$	$43.5$
AerialFusionMapNet	AID	AID	$8.1$	$3.8$	$4.3$	$4.4$	$0.3$	$9.5$	$4.3$	$0.4$

Table 4: Online HD map construction performance on the full 53-class KITScenes label space, evaluated within a

60\text{\times}30\text{\,}\mathrm{m}

region of interest. For readability, class-wise AP is grouped into six semantic categories: Lane Markings (LM), Lane Centerlines (LC), Road Infrastructure (RI), Traffic Lights (TL), Traffic Signs (TS), and Road Markings (RM). We additionally report the KITScenes topology score (Topo). C: camera images, AID: aerial image data.

Method	Input	mAP 53	$\mathrm{AP}_{\mathrm{Topo}}$
StreamMapNet [ 43 ]	C	$4.2$	$17.7$
StreamMapNet-CVS [ 20 ]	C + AID *	$4.9$	$23.7$
AerialFusionMapNet [ 19 ]	C + AID	$10.4$	$32.2$

Table 5: Extended-range online HD map construction performance on the full 53-class KITScenes label space using a

100\text{\times}50\text{\,}\mathrm{m}

region of interest. We report mean average precision and the KITScenes topology score. C: camera images, AID: aerial image data. * AID only used during training.

	$60\text{\times}30\text{\,}\mathrm{m}$		$100\text{\times}50\text{\,}\mathrm{m}$
Aerial GSD	mAP 53	$\mathrm{AP}_{\mathrm{Topo}}$	mAP 53	$\mathrm{AP}_{\mathrm{Topo}}$
$20\text{\,}\mathrm{cm}\text{/}\mathrm{px}$	$18.0$	$43.5$	$10.4$	$32.2$
$8\text{\,}\mathrm{cm}\text{/}\mathrm{px}$	$22.3$	$44.2$	$17.2$	$28.8$

Table 6: Impact of aerial image resolution on AerialFusionMapNet in the 53-class KITScenes setting across two regions of interest. We report mAP 53 and the KITScenes topology metric

\mathrm{AP}_{\mathrm{Topo}}

			$60\text{\times}30\text{\,}\mathrm{m}$				$100\text{\times}50\text{\,}\mathrm{m}$
Method	Training	Inference	AP ${}_{\text{ped}}$	AP ${}_{\text{div}}$	AP ${}_{\text{bnd}}$	mAP	AP ${}_{\text{ped}}$	AP ${}_{\text{div}}$	AP ${}_{\text{bnd}}$	mAP
StreamMapNet [ 43 ]	C	C	10.8	30.3	31.1	24.1	7.2	15.5	12.5	11.7
StreamMapNet- CVS [ 20 ]	C + AID	C	11.8	30.3	30.1	24.1	8.1	17.6	13.1	12.9
AerialFusionMapNet [ 19 ]	C + AID	C + AID	27.3	45.4	36.6	36.5	29.5	33.2	22.8	28.5

Table 7: Online HD map construction results for the previously established three-class protocol. C: camera images, AID: aerial image data.

Panel of Figure 4 — Figure 4: Rasterized HD map segmentation results by a DPT [ 33 ] decoder trained and inferred directly on the released DINOv3 features. Left to right: ground truth, prediction, higher-resolution aerial imagery (DINOv3 input), PCA of DINOv3 features.

Panel of Figure 5 — Figure 5: OpenSatMap [ 44 ] HD map segmentation representation and qualitative segmentation results. Although OpenSatMap closely relates to our aerial HD map construction task, the map representation, aerial image quality, and benchmark task drastically differ from ours. Left to right: ground truth, prediction, aerial image. Plots and image from [ 44 ] .

Method	Evaluation Data	Semantic Classes	mIoU
SegNeXt-T [ 10 ]	OpenSatMap19 [ 44 ]	8 line classes	28.71
SegNeXt-T [ 10 ]	OpenSatMap20 [ 44 ]	8 line classes	33.69
DINOv3 SAT ViT-L [ 36 ] + DPT [ 33 ]	KITScenes Aerial DINOv3 val	native 53 HD map classes	8.39
DINOv3 SAT ViT-L [ 36 ] + DPT [ 33 ]	KITScenes Aerial DINOv3 val	8 line classes (remapped * to [ 44 ] )	35.01

Table 8: Comparison of semantic HD map segmentation (mIoU in %) on KITScenes Aerial versus OpenSatMap. We compare a SegNeXt-tiny on the OpenSatMap val split (8 line classes: background, curb, virtual, and five painted line types) against a frozen DINOv3 SAT ViT-L with a DPT head trained on 53 Lanelet2 classes. On KITScenes, we report all native 53 classes as well as classes remapped * to the 8-class OpenSatMap ontology.

	Trans. [m] $\downarrow$		Orient. [ ∘ ] $\downarrow$
Setting	Mean	Median	Mean	Median
KITTI	7.45	4.03	3.33	1.88
KITScenes (zero-shot)	11.03	8.87	4.85	3.39
KITScenes (fine-tuned)	5.13	3.11	2.81	2.03

Table 9: FG 2 [ 41 ] cross-view localization results under cross-area evaluation, reporting translational and orientation error on the val split. KITTI denotes the reported result from [ 41 ] . For KITScenes Aerial, zero-shot uses the KITTI-trained model without adaptation, while fine-tuned denotes the model adapted on the KITScenes training split.

Appendix figures & tables10 assets

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 15 — Figure A.1 : Additional KITScenes Aerial examples showing synchronized multimodal observations, HD map annotations, and corresponding aerial representations.

Panel of Figure 16 — Figure A.2 : Additional KITScenes Aerial examples across further scene configurations.

Panel of Figure 17 — Figure B.3 : Qualitative comparison of country-level open aerial imagery (left) and high-quality municipal aerial imagery (right). We provide encoded neural features for higher-resolution aerial images of municipal quality. Both images shown here are at a resolution of $20\text{\,}\mathrm{cm}\text{/}\mathrm{px}$ due to privacy regulations.

Lane Markings (LM)	Lane Centerlines (LC)	Road Infrastructure (RI)
Road border Dashed Solid Solid Solid Solid dashed Dashed solid Bike dashed Bike solid Pedestrian crossing Zebra crossing Stop line	Centerline Bike centerline	Curbstone high Curbstone low Fence Building Wall Guardrail Drivable area Divider

Appendix

Table B.1 : Taxonomy of map elements used in the Lanelet2 map, grouped into six categories: Lane Markings (LM), Lane Centerlines (LC), Road Infrastructure (RI), Traffic Lights (TL), Traffic Signs (TS), and Road Markings (RM).

Subgroup	[email protected]			[email protected]			[email protected]
Subgroup	20 cm	8 cm	$\Delta$	20 cm	8 cm	$\Delta$	20 cm	8 cm	$\Delta$
Lane Markings (LM)	17.8	22.0	+4.2	29.2	32.0	+2.7	35.6	37.9	+2.4
Lane Centerlines (LC)	19.6	23.3	+3.6	31.5	32.2	+0.8	36.4	36.9	+0.5
Road Infrastructure (RI)	13.0	16.0	+3.0	23.6	26.5	+2.9	29.8	32.2	+2.5
Traffic Lights (TL)	8.7	14.8	+6.0	20.3	23.0	+2.7	26.2	26.9	+0.7
Traffic Signs (TS)	2.6	3.2	+0.6	4.1	5.1	+1.0	4.9	6.2	+1.2

Appendix

Table C.2 : AP comparison between 20 cm and 8 cm GSD across the six KITScenes reporting categories and different distance thresholds for the

60\text{\times}30\text{\,}\mathrm{m}

region of interest. Values are reported in %.

\Delta

denotes the absolute gain of 8 cm over 20 cm GSD in percentage points. The category grouping follows KITScenes; class assignments are shown in Table B.1 .

ROI	AP LM	AP LC	AP RI	AP TL	AP TS	AP RM	mAP 53	$\mathrm{AP}_{\mathrm{Topo}}$
$60\text{\times}30\text{\,}\mathrm{m}$	$30.6$	$30.8$	$24.9$	$21.5$	$4.8$	$48.9$	$22.3$	$44.2$
$100\text{\times}50\text{\,}\mathrm{m}$	$24.0$	$22.8$	$20.3$	$16.8$	$2.4$	$39.3$	$17.2$	$28.8$

Appendix

Table C.3 : Detailed results for AerialFusionMapNet using the higher-resolution

8\text{\,}\mathrm{cm}\text{/}\mathrm{px}

aerial imagery across the six KITScenes reporting categories and two regions of interest.

Figure D.4 : Frankfurt KITScenes rasterized HD Maps, GT Stitched

Figure D.5 : Frankfurt Validation DINOv3 Features + DPT Head Predictions Stitched

Figure D.6 : Karlsruhe KITScenes rasterized HD Maps, GT Stitched

Figure D.7 : Karlsruhe Validation DINOv3 Features + DPT Head Predictions Stitched

Autonomous Driving Benchmarks

Autonomous driving benchmarks evaluate how well vehicles perceive, predict, plan, and act safely across varied roads and conditions.

See all 141 papers

AerialFusionMapNet: Online HD Map Construction with Aerial-Onboard BEV Fusion
Jun 23, 2026Daniel Lengerer, Mathias Pechinger, Klaus Bogenberger +1Simultaneous Localization and Mapping HD Map Construction
Technical University of Applied Sciences Augsburg, Germany. · Technical University of Munich, Germany.
The Road Ahead in Autonomous Driving: The KITScenes Multimodal Dataset
Jun 1, 2026Richard Schwarzkopf, Fabian Immel, Alexander Blumberg +21Autonomous Driving Datasets Autonomous Driving Benchmarks
FZI Research Center for Information Technology · Karlsruhe Institute of Technology · University Charles III of Madrid +1
HRDX: A Large-Scale Vector HD-Map Dataset
Jun 11, 2026Sahith Reddy Chada, Isht Dwivedi, Nirav SavaliyaAutonomous Driving Datasets HD Map Construction
Honda Research Institute US, San Jose, CA.

Jun 23, 2026J/K move · Enter open · S save

AerialFusionMapNet: Online HD Map Construction with Aerial-Onboard BEV Fusion

Daniel Lengerer, Mathias Pechinger, Klaus Bogenberger, Carsten Markgraf

Technical University of Applied Sciences Augsburg, Germany. · Technical University of Munich, Germany.

Simultaneous Localization and Mapping HD Map Construction Autonomous Driving Perception Bird's-Eye-View Representation Multimodal Sensor Fusion

Open paper AskPDF arXiv

High-resolution aerial imagery has recently emerged as a complementary modality for automated driving perception and has shown potential to improve birds-eye-view (BEV) scene understanding when fused with onboard sensors. Prior work demonstrated performance gains for online high-definition (HD) map construction through aerial-onboard fusion; however, conventional end-to-end fusion does not fully exploit the structural information contained in aerial representations. In this work, we introduce AerialFusionMapNet, a fusion-based mapping framework with a structured two-stage training strategy that explicitly enhances the contribution of aerial features within a unified pipeline. The proposed training scheme enables more effective integration of structural aerial priors. On the nuScenes geographic split, AerialFusionMapNet achieves up to 54.7 mAP, improving over prior aerial-onboard fusion baselines from 48.8 mAP by +5.9 absolute and +12.1% relative. The results suggest that structured training design, rather than increased architectural complexity, plays a more decisive role in unlocking the full potential of aerial imagery for online HD map construction. Code and trained models are available at https://github.com/DriverlessMobility/AerialFusionMapNet.

Abstract

Autonomous Driving Benchmarks

Explore similar work

AerialFusionMapNet: Online HD Map Construction with Aerial-Onboard BEV Fusion

The Road Ahead in Autonomous Driving: The KITScenes Multimodal Dataset

HRDX: A Large-Scale Vector HD-Map Dataset