believedCG

believedCG is an open-source software framework for studying perceived believability of computer-generated imagery (CGI). It provides a modular pipeline for scene generation, rendering, and user-study collection, enabling researchers to quantify how visual and contextual factors influence realism as perceived by observers.

The framework offers components for scene construction, a rendering-agnostic interface, and a believability modeling layer that maps participant responses to composite believability scores. It supports scripting in Python and offers hooks for integration with common rendering environments, allowing researchers to vary lighting, texture, motion, and scene context.

believedCG is designed for use in perceptual science, computer graphics research, and human-computer interaction studies. Its architecture emphasizes reproducibility, with configuration files that capture experiment design, stimuli, and evaluation metrics, and with facilities to share datasets and analysis pipelines.

Development began in the mid-2020s through a collaboration of researchers and independent developers. The project maintains a public repository and community guidelines to encourage collaboration, documentation, and example studies.

In practice, believedCG has been used to compare believability across rendering settings and animation styles, and to teach concepts of realism in CG. Critics note the reliance on subjective ratings and the challenge of cross-study comparability, motivating ongoing work on standardized benchmarks and objective correlates.