Data and Code for Generative AI-Driven Window View Design and Evaluation: A Multi-Sensory Framework with Multi-Criteria Decision Integration

This repository provides research data and supporting computational materials associated with the study entitled "Generative AI-Driven Window View Design and Evaluation: A Multi-Sensory Framework with Multi-Criteria Decision Integration." The study investigates visually evoked multisensory expectations in window-view design and develops a computational framework integrating human sensory judgements, interpretable spatial annotations, multicriteria decision analysis, and generative artificial intelligence. The research involved 1,000 paired window-view images and sensory annotation maps, questionnaire-based evaluations from 523 participants, and the assessment of 200 AI-generated window-view design proposals. Four sensory dimensions were considered: auditory, olfactory, tactile, and gustatory perception. The deposited materials document the datasets and analysis outputs that can be shared under applicable ethical, privacy, and copyright requirements. These include the releasable image–annotation pairs, processed questionnaire-derived data, Pix2PixGAN validation and prediction results, human–AI score comparisons, and supporting training records and source code, as specified in the accompanying README. The original study used a training, validation, and test partition of 600, 200, and 200 paired images, respectively. The README documents the released data structure, file-level provenance, preprocessing procedures, and limitations relevant to reproducibility. Importantly, the sensory ratings represent cross-modal expectations elicited by visual stimuli rather than direct measurements of acoustic, olfactory, tactile, or gustatory conditions in physical environments. Funding: This study was supported by the Humanities and Social Sciences Fund Project of the Ministry of Education of China (Grant No. 25YJCZH010). Contact: Shaoting Zeng, Beijing University of Technology, [e-mail].