SA BENCH SA BENCH is the benchmark dataset released with “Beyond Pixels: Benchmarking and Reward Based Assessing Framework for Visual Spatial Aesthetics.” It evaluates the spatial aesthetics of interior images along four dimensions: distortion harmony layout lighting The dataset contains 17,768 annotated examples across four spatial aesthetic dimensions, with image assets and human annotations for training and evaluation. Dataset Details Dataset Description SA BENCH is designed for image quality and aesthetic assessment of interior scenes. It focuses on spatial aesthetics rather than generic image appeal, and provides dimension specific annotations for: distortion : geometric distortion, deformation, alignment errors, and material realism harmony : style consistency, color coordination, and overall visual coherence layout : spatial arrangement, balance, and positional relationships of key elements lighting : illumination quality, shadow realism, light source consistency, and atmosphere Dataset Sources Repository: this Hugging Face dataset repository Associated paper: Beyond Pixels: Benchmarking and Reward Based Assessing Framework for Visual Spatial Aesthetics Code and model releas…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy