The Spectrascapes Dataset: Street-view imagery beyond the visible captured using a mobile platform

Akshit Gupta; Joris Timmermans; Filip Biljecki; Remko Uijlenhoet

cs.CVApr 14, 2026

The Spectrascapes Dataset: Street-view imagery beyond the visible captured using a mobile platform

Authors: Akshit Gupta, Joris Timmermans, Filip Biljecki, Remko Uijlenhoet

Organizations: TU Delft, Stevinweg 1, Delft, 2628CN, Zuid-Holland, The Netherlands. · National University of Singapore, 4 Architecture Dr, Singapore, 117566, Singapore.

Abstract

High-resolution data in spatial and temporal contexts is imperative for developing climate resilient cities. Current datasets for monitoring urban parameters are developed primarily using manual inspections, embedded-sensing, remote sensing, or standard street-view imagery (RGB). These methods and datasets are often constrained respectively by poor scalability, inconsistent spatio-temporal resolutions, overhead views or low spectral information. We present a novel method and its open implementation: a multi-spectral terrestrial-view dataset that circumvents these limitations. This dataset consists of 17,718 street level multi-spectral images captured with RGB, Near-infrared, and Thermal imaging sensors on bikes, across diverse urban morphologies (village, town, small city, and big urban area) in the Netherlands. Strict emphasis is put on data calibration and quality while also providing the details of our data collection methodology (including the hardware and software details). To the best of our knowledge, Spectrascapes is the first open-access dataset of its kind. Finally, we demonstrate two downstream use-cases enabled using this dataset and provide potential research directions in the machine learning, urban planning and remote sensing domains.

Figures & tables

Work	Morphological Diversity	Spectral Bands in Imagery			Custom Platform	Open Access
		RGB	NIR	Thermal
Klimenka et al. (2025)	Limited (1 city)	✗	✗	✓	No	No
Gupta et al. (2024b)	Limited (1 city)	✓	✓	✗	Yes	No
Tarkhan et al. (2025)	Limited (3 metro. hubs)	✓	✗	✗	No	Yes
Hou et al. (2024)	Diverse	✓	✗	✗	No	Yes
This work	Diverse (4 morphologies)	✓	✓	✓	Yes	Yes

Table 1: Comparison of recent street-view imagery datasets by spatial coverage, spectral richness, and open accessibility, and our contribution.

Figure 1: Overall summary of our approach for data collection, using bikes to capture multi-spectral terrestrial imagery data.

Figure 2: Architecture of our system. Refer to Table 2 for details of each imaging sensor.

Imaging Sensor	Sensor Name	Spectral Bands	Image Size (pixels)	Manufacturer (Accuracy)	Connection Signals (from control unit)	High-cost substitution
RGN	Mapir Survey 3W RGN MAPIR Camera (2023b)	500-575nm (Green) 630-690nm (Red) 810-890nm (N-Infrared)	4000x3000	$<1\%$ (Lens)	PWM	Micasense Altum-PT MicaSense (2022)
Thermal	Flir Lepton 3.5 Teledyne FLIR (2021)	8,000nm-14,000nm (Long-Wave Infrared)	160x120	$\pm 5^{\circ}\text{C}$ or $5\%$	RPC over UART	Flir Boson Teledyne FLIR (2023)
RGB	Pi Camera Module 3 Sony IMX708 Raspberry Pi Ltd (2023)	480–600nm (Green) 590-700nm (Red) 400-540nm (Blue)	4608×2592	N/A	I2C and MIPI CSI-2	Multiple Available

Table 2: Details of the imaging sensors utilized in the hardware platform.

Panel of Figure 3 — Figure 3: Hardware and its deployment on a rental bike (OV-fiets bicycle) in the Netherlands.

Figure 4: The study areas in the Netherlands. The blue dots indicate the locations where the multi-spectral imaging data was gathered with a bicycle. Arnhem is a large city with a population of around 167,632 people, Delft is a small city with a population of around 109,577 people, Oirschot is a town with a population of 19,280 people and Middelbeers is a small village with a population of 4,994 people. All maps presented with a north-up orientation. Population numbers sourced from Netherlands (2025) .

Panel of Figure 5 — Figure 5: Example of the multi-spectral imaging data, and the remote sensing image obtained using the metadata (Lat: 51.986186667, Long: 5.904403333) at each data observation location. This data was captured on Zijpendaalseweg street in Arnhem.

Panel of Figure 6 — Figure 6: Calibration plates utilized for intrinsic calibration and extrinsic calibration of all the imaging sensors, and spectral calibration of the RGN imaging sensor.

Calibration Type	Individual Sensor / Sensor Pair	RMS Error (pixels)
Intrinsic	RGB	0.49
	Thermal	0.35
	RGNear-Infrared	0.58
Extrinsic (Static)	RGB $\leftrightarrow$ Thermal	0.61
	RGB $\leftrightarrow$ RGNear-Infrared	0.54

Table 3: Intrinsic and extrinsic calibration re-projection errors for the imaging sensors to understand the internal characteristics of the imaging sensors and estimating the physical geometry of our data collection hardware respectively.

Urban morphology	Total number of images	Images with valid GPS coordinates
Arnhem (Big urban area)	7,065	6,918
Delft (Small city)	3,168	1,851
Oirschot (Town)	5,595	4,974
Middelbeers (Village)	1,890	1,833
Total	17,718	15,576

Table 4: Distribution of the total number of images for different urban morphologies with valid geospatial location

Panel of Figure 7 — Figure 7: Depth map of a street in Arnhem using the VGGT model Wang et al. (2025) . The red, green, and blue cones denote the estimated positions of the imaging sensor along the bike’s trajectory where the images were acquired.

Panel of Figure 8 — Figure 8: A comparison of building façades from a single vantage point. The near-infrared channel analysis shows that the off-white materials on the two façades, which appear visually similar in the RGB images, are markedly different. Additionally, notice how the temperature of façades are different. (a) Standard RGB (Red, Green, Blue) image with only visible spectra. (b) Thermal image from the same vantage point, where darker pixels indicate lower temperature. (c) RGN (Red, Green, Near-Infrared) image at the same vantage point. (d) Zoom-in view of the RGB image showing how the off-white material on both façades appears visually identical. (e) Zoom-in view of the RGN image with a heat-map applied to the near-infrared channel, demonstrating a marked difference between the off-white materials. (f) Box-plot after selecting 10 different regions-of-interest (ROI)/group of pixels for the off-white material in left façade, the off-white material in right façade and the glass material in the right façade. The y-axis indicates the average near-infrared reflectance for the regions-of-interest in digital format ranging from 0-255.

Remote Sensing

Remote sensing uses satellite, aerial, and ground-based observations to measure and map Earth and other planetary surfaces.

See all 528 papers

Urban Heat MiniCubes: An AI-Ready dataset for urban heat research
Jun 10, 2026Jonathan Starfeldt, Maria J. Molina, Alexander Kerr +3Remote Sensing
Department of Atmospheric and Oceanic Science, University of Maryland, College Park, MD, USA. · Department of Computer Science, University of Maryland, College Park, MD, USA. · NASA Goddard Space Flight Center, Greenbelt, MD, USA. +1
BELDE: Building a Large-scale Earth-observation Land-cover Dataset for Europe
Jun 18, 2026Ümit Mert Çağlar, Alptekin TemizelRemote Sensing Image Segmentation Distribution Shift
Graduate School of Informatics, Middle East Technical University, Ankara, Turkey
Designing streetscapes from street-view imagery using diffusion models
May 17, 2026Yuzhou Chen, Yuebing Liang, Lingqian Hu +4Image Generation Diffusion-Based Image Generation
Department of Urban and Regional Planning, University of Florida · Singapore-MIT Alliance for Research and Technology Centre (SMART) · Department of Landscape Architecture and Urban Planning, Texas A&M University +1

Jun 10, 2026J/K move · Enter open · S save

Urban Heat MiniCubes: An AI-Ready dataset for urban heat research

Jonathan Starfeldt, Maria J. Molina, Alexander Kerr, Adam Yang+2

Department of Atmospheric and Oceanic Science, University of Maryland, College Park, MD, USA. · Department of Computer Science, University of Maryland, College Park, MD, USA. · NASA Goddard Space Flight Center, Greenbelt, MD, USA.+1

Remote Sensing

Open paper AskPDF arXiv

Urban heat is amplified by impermeable surfaces and heterogeneous built environments, yet street-level variability remains difficult to quantify because multi-sensor observations are rarely available in consistent, analysis-ready form at the necessary spatiotemporal scales. We present "Urban Heat MiniCubes," a publicly available, FAIR-oriented dataset designed for machine learning applications in urban heat research. The dataset provides harmonized 90 x 90 km gridded data cubes for 48 cities in the Western Hemisphere spanning 2022-2023, with variables reprojected and collocated to a common grid to reduce preprocessing (e.g., reprojection, resampling, and spatiotemporal alignment). Urban Heat MiniCubes includes two complementary modalities: (i) higher-spatial-resolution, lower-frequency observations from Landsat 8/9 (e.g., surface reflectances) and Sentinel-1 (e.g., synthetic aperture radar backscatter), and (ii) higher-temporal-frequency, coarser observations from GOES-R (e.g., longwave infrared brightness temperatures) and a microwave land surface temperature product. We document variables and metadata and provide technical assessment using inter-variable analyses and autoencoder-based reconstruction-error summaries across pixel classes (e.g., water and cloud). Potential use cases and limitations are also discussed.

Abstract

Remote Sensing

Explore similar work

Urban Heat MiniCubes: An AI-Ready dataset for urban heat research

BELDE: Building a Large-scale Earth-observation Land-cover Dataset for Europe

Designing streetscapes from street-view imagery using diffusion models