The purpose of this paper is to explore the public's perceived restorative to urban green spaces (UGS). Compared to Modern Urban Park (MUP), little research has been done on the restoration potential of Classical Chinese Garden (CCG). To fill this gap, we collect video clips of various scenarios to produce video images of CCG and MUP representing UGS. Public evaluations of the videos were collected, and the data were analyzed by combining the Short Version Revision Repair Scale (SRRS) and deep learning techniques. They are finding proof for enhancing the healing effects of CCG and MUP. The results indicated that the difference between the two restorations was not significant. Videos representing CCG had a greater standard deviation of restorative (lower consensus) compared to videos depicting MUP. Deep learning techniques for semantic segmentation combined with expert scoring methods can effectively help us to understand the drivers affecting resilience, and we combine our findings to conclude that improved water feature design is an essential driver for enhancing the perceived restorative of CCGs and that decreasing specific artificial modern structures, enhancing vegetation cover, and increasing public exposure to nature is critical to strengthening the restorative of both. Hopefully, these findings will improve visitors' recovery in UGS environments and guide landscape architects to intervene more effectively in healing design in UGS.