The vertical developments in cities reshape the urban form and structure, and the influences on human liveability can be reflected by the variations in property values. The hedonic price model (HPM) is commonly employed in city-scale property valuation to unravel the hedonic values of different influential variables. In vertically developed cities, it necessitates the exploration of the hedonic value in the vertical dimension (3D), which was previously under-researched due to limited 3D data and the complexity of processing techniques. Recent studies use eye-level street view images (SVIs) for valuation, but the 3D perspective is still missing. This study proposes a novel 3D property valuation method using SVIs acquired from two angles, eye-level (pitch 0°) and sky-view (pitch 90°, upwards), and machine learning method to complete the 3D perspective and provide explainability of 3D HPM. We also compared different valuation models – namely Ordinary Least Square (OLS), Geographically Weighted Regression (GWR), and Random Forest (RF) – using model performance metrics. Our main findings include: 1) 3D variables are statistically significant, and adding them improves the model performance (R2 from 0.580 to 0.636 in GWR); 2) In the sky-view angle, the proportion of sky has a positive correlation while the presence of buildings and trees are negatively correlated with property values; 3) RF outperforms OLS and GWR with the highest R2 (0.768) and the least RMSE (1669.60 yuan/m2), which demonstrates its robust explainability and applicability for valuation. This study enriches the property valuation literature on the significance of the 3D variables and provides references to guide fair taxation and equal land use policy in vertically developed cities.