# RECORD 5bcd886348a12f52aa055354a9647ba5c4fa64c4 # TITLE: Giving Eyes to Automated Valuation Models: Computer Vision Assessment of Interior Home Condition # YEAR: 2026.0 VENUE: Journal of real estate finance and economics DOI: 10.1007/s11146-026-10071-w # AUTHORS: V. Almaas; A. Oust; A. J. Pollestad # PAGES: 35 CANDIDATE TABLE PAGES (1-based): [19, 21, 22, 31] # ABSTRACT: Housing condition is a key determinant of property value, yet it is often omitted from modern automated valuation models (AVMs). This study develops a computer vision framework to extract room-specific condition signals from listing images and examines their economic relevance for housing valuation. Using 15,702 condominium transactions from Oslo, Norway (2022–2023), matched with 393,914 images, we classify images by room type, estimate condition on a 1–5 scale, and incorporate these measures into hedonic linear regression and XGBoost AVMs. The results show that computer-vision-based condition improves predictive performance, with gains comparable to those from human-labeled condition. Moreover, housing condition is positively associated with sales prices, with stronger effects for kitchens and bathrooms and in lower-priced homes. Explainable AI analysis further indicates that the models rely on intuitive and economically meaningful visual patterns. ===== PAGE 1 (score 7, numeric-paren 0, table-words 0, var-words 7) ===== The Journal of Real Estate Finance and Economics https://doi.org/10.1007/s11146-026-10071-w Arne Johan Pollestad arne.j.pollestad@ntnu.no Vebjørn Almaas vebjorn.almaas@ntnu.com Are Oust are.oust@ntnu.no 1 NTNU Business School, Norwegian University of Science and Technology, Trondheim, Norway 2 Norwegian School of Economics, Bergen, Norway Received: 23 September 2025 / Accepted: 26 June 2026 © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2026 Giving Eyes to Automated Valuation Models: Computer Vision Assessment of Interior Home Condition Vebjørn Almaas1,2  · Are Oust1  · Arne Johan Pollestad1 Abstract Housing condition is a key determinant of property value, yet it is often omitted from modern automated valuation models (AVMs). This study develops a computer vision framework to extract room-specific condition signals from listing images and examines their economic relevance for housing valuation. Using 15,702 condomin­ ium transactions from Oslo, Norway (2022–2023), matched with 393,914 images, we classify images by room type, estimate condition on a 1–5 scale, and incorpo­ rate these measures into hedonic linear regression and XGBoost AVMs. The results show that computer-vision-based condition improves predictive performance, with gains comparable to those from human-labeled condition. Moreover, housing condi­ tion is positively associated with sales prices, with stronger effects for kitchens and bathrooms and in lower-priced homes. Explainable AI analysis further indicates that the models rely on intuitive and economically meaningful visual patterns. Keywords  Automated valuation model (AVM) · Image segmentation · Condition · Computer vision · Explainable AI · Real estate valuation JEL-Classification  C55 · R21 · R31 ===== PAGE 19 (score 11, numeric-paren 0, table-words 3, var-words 3) ===== 1 3 Giving Eyes to Automated Valuation Models: Computer Vision… Table 4  AVM performances with condition variables AVM Condition variable MAPE MdAPE R2 > 10% HLR Excluding condition 12.75% 10.11% 0.672 49.70% HLR Computer vision condition 12.42% 9.91% 0.687 50.55% HLR Human condition 12.41% 9.75% 0.690 51.19% XGBoost Excluding condition 8.03% 5.76% 0.847 73.43% XGBoost Computer vision condition 7.71% 5.49% 0.855 75.68% XGBoost Human condition 7.49% 5.34% 0.863 75.76% This table shows the predictive performances of the two AVMs, a hedonic linear regression (HLR) and an XGBoost, under three different specifications. First, we train our model only on traditional housing data. Secondly, we include thehi computer vision condition variable to the input data. Finally, we replace the computer vision condition variable with human condition labels from the crowdsourcing process. The best-performing model(s) for each AVM is highlighted in bold. Source: Author calculations using Boligmappa AS housing data, January 2022–December 2023 Fig. 7  Average condition levels across city districts. Notes: This figure displays the spatial distribution of average room condition across the administrative districts of Oslo. A blue gradient with a corre­ sponding numerical scale is adopted to represent the average condition levels, where lighter (darker) shades indicate worse (better) condition. The 15 district names are labeled on the map. Oslo Sentrum (CBD) is marked in gray due to the absence of data in our sample. Source: Author calculations using Boligmappa AS housing data, January 2022–December 2023 ===== PAGE 20 (score 11, numeric-paren 0, table-words 2, var-words 7) ===== 1 3 V. Almaas et al. economically meaningful information that is not fully captured by standard housing characteristics. For the HLR model, MAPE declines from 12.8% to 12.4% when computer- vision-based condition is included, while R2 increases from 67.2% to 68.7%. These improvements suggest that the condition measures capture variation in housing qual­ ity that is priced in the market but omitted from conventional hedonic specifications. Importantly, the performance gains are similar when using human-labeled condition, indicating that the computer vision measures recover much of the same information. A similar pattern is visible for the XGBoost model. Including computer-vision- based condition reduces MAPE from 8.0% to 7.7%, while the addition of human con­ dition labels yields a further improvement (MAPE of 7.5%). Across the two models, this suggests that the information extracted by the computer vision models resembles that provided by human assessments of interior condition. A reduction in average valuation error of 0.4% points equals a reduction of around NOK 24,000 for the average home with a price per m2 of NOK 90,000 and a size of 66 m2. By extracting valuation-relevant information from listing images, computer vision models may therefore help reduce omitted variable bias in pricing models and improve the accuracy of housing valuations at an economically meaningful scale. Hedonic Pricing of Condition Signals Furthermore, we can examine the estimated coefficients of the condition variables in the HLR model, reported in Table 5. Columns 1–2 include city district fixed effects, while columns 3–4 employ more granular postal code fixed effects. Moving to finer spatial fixed effects allows us to absorb increasingly detailed location-specific price variation, thereby reducing concerns that the estimated condition effects are driven by unobserved neighborhood characteristics. The coefficients are thus identified from within-location variation in condition across otherwise comparable properties, pro­ viding a cleaner estimate of how computer-vision-based signals of interior condition are associated with predicted sales prices. With few exceptions, primarily insignificant estimates for low-condition bed­ rooms, the results consistently indicate that higher condition levels are associated with higher property prices. Economically, the effects are substantial, as a one-level increase in average condition is associated with a price premium of approximately 8.2% to 8.7%. This suggests that visual interior condition signals capture a meaning­ ful dimension of housing quality that adds value to price models. The stability of the condition coefficients across the different specifications also suggests that the results are not driven by spatial sorting of higher-condition homes into more expensive areas. To illustrate the magnitude of the effect, consider a representative condominium in our sample, with an average size of 66 m² and a price of approximately NOK 90,000 per square meter, corresponding to a total value of roughly NOK 5.9 million. An 8% increase in price therefore implies an increase of almost NOK 500,000. These esti­ mates highlight that relatively small improvements in measured condition translate into economically significant differences in predicted values. The room-specific condition variables further allow us to decompose the average effect across different parts of the dwelling. For bathrooms, condition is statistically ===== PAGE 21 (score 46, numeric-paren 27, table-words 5, var-words 6) ===== 1 3 Giving Eyes to Automated Valuation Models: Computer Vision… distinguishable from the reference category (level 4) across all levels, with clear val­ uation penalties for lower condition and premiums for higher ones. A similar pattern emerges for kitchens and living rooms, with the effects being particularly pronounced for kitchens. This suggests that kitchens may be the room type for which the image- based measures contain the most valuation-relevant information. Economically, the magnitudes are substantial as a level 1 kitchen is associated with a valuation discount of approximately 17.2% to 18.1%, while a level 5 kitchen commands a premium of 7.8% to 7.9% relative to the reference level. The results generally imply that visual condition measures capture meaningful heterogeneity in interior condition across room types. Next, we conduct a series of heterogeneity analyses to examine whether the results reported in Table 5 vary across subsamples of condominiums. First, we split the sam­ ple into quartiles based on transaction prices, comparing the lowest and highest price segments. This allows us to assess whether the valuation impact of image-based con­ Table 5  Hedonic linear regression condition effects Condition variable (1) (2) (3) (4) Room-specific Average Room-specific Average Dependent variable in line-eq-IEq12   in line-eq-IEq13 in line-eq-IEq14 in line-eq-IEq15 CondAverage 0.087*** 0.082*** CondBath = 1 -0.062** -0.057** CondBath = 2 -0.050*** -0.041*** CondBath = 3 -0.011*** -0.014*** CondBath = 5 0.027*** 0.025*** CondBed = 1 -0.099*** -0.087*** CondBed = 2 -0.043*** -0.031*** CondBed = 3 -0.004 -0.003 CondBed = 5 0.136 0.145* CondLiv = 1 --- --- CondLiv = 2 -0.045** -0.039** CondLiv = 3 -0.011*** -0.010*** CondLiv = 5 0.021 -0.015 CondKit = 1 -0.172*** -0.181*** CondKit = 2 -0.087*** -0.091*** CondKit = 3 -0.036*** -0.037*** CondKit = 5 0.079*** 0.078*** Dwelling Attributes Yes Yes Yes Yes Year x Quarter FE Yes Yes Yes Yes District FE Yes Yes No No Postal code FE No No Yes Yes Adjusted R2 0.6981 0.6952 0.772 0.771 Observations 15,702 12,084 15,702 12,084 The coefficients represent the percentage effect on price per m2 from a hedonic linear regression model for different condition variables. CondAverage refers to an aggregated condition variable across room types, while CondBath, CondBed, CondLiv, and CondKit represent the room-specific conditions. All conditions are measured on a scale from 1–5. The models have fixed effects for the interaction between year and quarter, as well as city district fixed effects in models 1–2 and postal code fixed effects in models 3–4. *p<0.1; **p<0.05; ***p<0.01. Source: Author calculations using Boligmappa AS housing data, January 2022–December 2023 ===== PAGE 22 (score 68, numeric-paren 41, table-words 7, var-words 5) ===== 1 3 V. Almaas et al. dition signals differs across the price distribution, for instance if interior quality is more strongly capitalized in high-end markets. Second, we split the sample based on the number of listing images, comparing the quartile with the fewest images to the quartile with the most. The number of images serves as a proxy for marketing effort and information availability, enabling us to test whether the valuation effects of con­ dition signals depend on the richness of visual information provided. Finally, we augment the baseline specification with broker fixed effects. Since listing images are produced by professional real estate brokers, this specification controls for systematic differences in photography quality, staging, and presenta­ tion across professionals. It thereby mitigates concerns that the estimated condition effects may partly reflect broker-specific practices correlated with both image quality and property values. The results are shown in Table 6. The table reports our preferred specification with postal code fixed effects, while the corresponding district fixed effects estimates are reported in Appendix Table 7. In the full-sample specification with broker fixed Table 6  Heterogeneity analysis of condition effects Sample (1) (2) (3) (4) (5) Cheap homes Expensive homes Few pictures Many pictures Full Dependent variable in line-eq-IEq16 in line-eq-IEq17 in line-eq-IEq18 in line-eq-IEq19 in line-eq-IEq20 CondBath = 1 -0.048* 0.051 -0.054* -0.188*** -0.068*** CondBath = 2 -0.030*** -0.01 -0.034*** -0.047*** -0.042*** CondBath = 3 -0.020*** 0.011** -0.010* -0.021*** -0.014*** CondBath = 5 0.024** 0.017 0.030** 0.007 0.023*** CondBed = 1 -0.053** --- -0.016 -0.157 -0.083*** CondBed = 2 -0.031*** 0.012 0.011 -0.047 -0.038*** CondBed = 3 -0.017*** 0.000 0.010* 0.003 -0.002 CondBed = 5 0.009 0.067 0.126 0.110 0.146* CondLiv = 1 --- --- --- --- --- CondLiv = 2 -0.022 -0.015 0.001 0.001 -0.036** CondLiv = 3 -0.015*** -0.007 -0.009 -0.009 -0.010*** CondLiv = 5 -0.057 0.020 0.034 0.017 -0.013 CondKit = 1 -0.079*** -0.170 -0.184*** -0.186 -0.148*** CondKit = 2 -0.059*** -0.025 -0.089*** -0.055* -0.077*** CondKit = 3 -0.021*** -0.009 -0.040*** -0.033*** -0.031*** CondKit = 5 -0.006 0.042*** 0.083** 0.061*** 0.076*** Dwelling Attributes Yes Yes Yes Yes Yes Year x Quarter FE Yes Yes Yes Yes Yes Postal code FE Yes Yes Yes Yes Yes Broker FE No No No No Yes Adjusted R2 0.714 0.283 0.861 0.694 0.756 Observations 3,926 3,926 3,110 3,105 15,702 The coefficients represent the percentage effect on price per m2 from a hedonic linear regression model for different condition variables. CondBath, CondBed, CondLiv, and CondKit represent the room- specific conditions. All conditions are measured on a scale from 1–5. The models have fixed effects for the interaction between year and quarter, as well as postal code fixed effects. The model in column 5 also includes broker fixed effects. *p<0.1; **p<0.05; ***p<0.01. Source: Author calculations using Boligmappa AS housing data, January 2022–December 2023 ===== PAGE 23 (score 13, numeric-paren 0, table-words 1, var-words 11) ===== 1 3 Giving Eyes to Automated Valuation Models: Computer Vision… effects in Column 5, level 1 kitchens are associated with a valuation discount of 14.8%, while level 2 and level 3 kitchens are discounted by 7.7% and 3.1%, respec­ tively. Bathrooms display a similar ordering, with lower condition levels associated with discounts and level 5 bathrooms associated with a positive premium. These patterns suggest that the image-based variables capture valuation-relevant signals of interior condition rather than merely reflecting broker-specific marketing practices. The heterogeneity analysis in Columns 1–4 further indicates that condition effects are more pronounced among lower-priced homes. In the cheapest quartile, poor kitchen, bathroom, and bedroom condition is consistently associated with lower prices, whereas the corresponding penalties are weaker and less precisely estimated among the most expensive homes. This suggests that visible interior deficiencies may be more informative for valuation in lower-priced market segments, where required renovation costs represent a larger share of the total property value. The results by number of listing images are more mixed. While some estimates suggest stronger penalties for poor condition among homes with richer visual docu­ mentation, these patterns are less robust once postal code fixed effects are included. We therefore interpret this evidence cautiously. The HLR coefficients from Table 5 align prior with AVM research that emphasize the importance of incorporating detailed, room-specific condition measures to explain sales prices (Koch et al., 2024; Miller et al., 2018). The strong price premiums for high-standard bathrooms and the penalties for substandard conditions supports the argument made by Oust et al. (2023) that interior quality significantly affects property valuation. Furthermore, the differential impact across price segments, where poor bathroom condition appears to reduce sale prices more for low-end properties, reso­ nates with the broader evidence from Mamre and Sommervoll (2024) and Francke and van de Minne (2017), who show that both renovations and lack of maintenance have substantial effects on sales prices. Non-Linear Model Interpretation and Explainability After assessing the hedonic pricing of condition signals, explainable AI methodology allows for similar analysis for the non-linear XGBoost AVM. These are displayed in Fig. 8. Both Panel A, which displays the four room-specific condition variables, and Panel B, which shows the average condition level, reveal a clear positive relationship between condition and property price. This is evident since higher condition values (in red colors) have positive marginal effects on predicted sales price (positive SHAP values). To further examine the marginal effect of condition, we present SHAP dependence plots in Panels A–E of Fig. 9. In these plots, the horizontal axis indicates condition level, the vertical axis shows the marginal price impact (SHAP value) of condition, and coloring reflects the actual sales price. Ignoring coloration for the moment, the relationship between condition and marginal price impact is consistently positive across all room types: higher condition levels lead to higher predicted prices from the XGBoost AVM, holding other variables constant. As indicated by the confusion matrices in Figs. 3, 4, 5, 6, and 15 in the Appendix, a large share of homes was classified at levels 3 and 4. This distribution explains why ===== PAGE 31 (score 61, numeric-paren 39, table-words 6, var-words 2) ===== 1 3 Giving Eyes to Automated Valuation Models: Computer Vision… However, an important difference is that the condition coefficients are more fre­ quently statistically significant in the subsamples based on the number of images. In particular, the estimates for homes with few and many pictures tend to be larger in magnitude and more precisely estimated under the district-level specification. This likely reflects the broader spatial controls, which leave more variation in prices to be explained by observable characteristics such as condition. The results suggest that while the main findings are robust to the choice of spatial fixed effects, the magni­ tude and statistical significance of the condition effects are somewhat sensitive to the level of spatial aggregation. The more conservative postal code specification in Table 6 absorbs finer location-specific price variation, reducing the risk that condi­ tion coefficients capture unobserved neighborhood characteristics. By contrast, the district-level specification in Table 7 may partially assign residual spatial variation to condition, leading to stronger and more precisely estimated effects. Importantly, the consistency in the direction and economic interpretation of the coef­ ficients across both specifications supports the conclusion that the computer-vision- based condition measures capture valuation-relevant information on interior quality, rather than merely reflecting spatial sorting or differences in marketing practices. Table 7  Heterogeneity analysis of condition effects Sample (1) (2) (3) (4) (5) Cheap homes Expensive homes Few pictures Many pictures Full Dependent variable in line-eq-IEq21   in line-eq-IEq22   in line-eq-IEq23   in line-eq-IEq24   in line-eq-IEq25 CondBath = 1 -0.034 0.010 -0.050 -0.218*** -0.062** CondBath = 2 -0.039*** -0.019 -0.043*** -0.051*** -0.049*** CondBath = 3 -0.007 0.01** -0.010* -0.022*** -0.012*** CondBath = 5 0.013 0.021* 0.036** 0.004 0.026*** CondBed = 1 -0.0504* --- -0.016 -0.222 -0.095*** CondBed = 2 0.035*** 0.022 0.010 -0.070* -0.041*** CondBed = 3 -0.023*** 0.002 0.009 -0.003 -0.003 CondBed = 5 0.035 0.042 0.119 0.132 0.146* CondLiv = 1 --- --- --- --- --- CondLiv = 2 -0.030 0.008 0.013 0.010 -0.044** CondLiv = 3 -0.013*** -0.008 -0.012** -0.004 -0.010*** CondLiv = 5 0.013 0.005 0.033 0.051 0.018 CondKit = 1 -0.083*** -0.157 -0.191*** -0.419 -0.168*** CondKit = 2 -0.059*** -0.024 -0.085*** -0.077** -0.083*** CondKit = 3 -0.022*** -0.012** -0.037*** -0.034*** -0.034*** CondKit = 5 -0.029 0.052*** 0.089 0.066*** 0.081*** Dwelling Attributes Yes Yes Yes Yes Yes Year x Quarter FE Yes Yes Yes Yes Yes District FE Yes Yes Yes Yes Yes Broker FE No No No No Yes Adjusted R2 0.608 0.141 0.825 0.592 0.707 Observations 3,926 3,926 3,110 3,105 15,702 The coefficients represent the percentage effect on price per m2 from a hedonic linear regression model for different condition variables. CondBath, CondBed, CondLiv, and CondKit represent the room-specific conditions. All conditions are measured on a scale from 1–5. The models have fixed effects for the interaction between year and quarter, as well as district fixed effects. The model in column 5 also includes broker fixed effects. *p<0.1; **p<0.05; ***p<0.01. Source: Author calculations using Boligmappa AS housing data, January 2022–December 2023 ===== PAGE 32 (score 3, numeric-paren 0, table-words 0, var-words 3) ===== 1 3 V. Almaas et al. Distribution of Classified Condition Levels Figure 15 summarizes the condition variables derived from the CNN computer vision classifications. Across all four room types, the majority of images are classified at condition levels 3 and 4, indicating generally high maintenance standards. Overall, condition levels 1, 2, and 5 are underrepresented, suggesting that most Oslo condo­ miniums fall within the mid- to high-quality range. Acknowledgements  We thank the participants of the American Real Estate Society (ARES) Spring Con­ ference 2026 for their valuable comments and discussions. We are also grateful to the two anonymous reviewers for their thoughtful and constructive feedback, which substantially improved the quality and clarity of the manuscript. An earlier version of this work was submitted as a master’s thesis by V.A. at the Norwegian School of Economics in the fall of 2024. The present article has undergone extensive revisions and extensions and should be regarded as an independent research contribution. Any remaining errors or omissions are solely the responsibility of the authors. Author Contributions  All authors contributed to the conceptualization and methodology. V.A. performed the data cleaning and data analysis. A.O. and A.P. wrote the main manuscript text. All authors participated in revising the manuscript and approved the final version. Fig. 15  Classification frequency. Notes: This figure shows the distributions of condition classification frequencies for four different room types. All condition assessments are classified using computer vi­ sion technology. Condition is assessed on a scale from 1–5, where higher numbers represent better con­ dition. Source: Author calculations using Boligmappa AS housing data, January 2022–December 2023