How I Perceive It

How I Perceive It challenges the assumption that AI image captioning and visual interpretation are purely model-driven processes, without incorporating human experience.

Everyone perceives the world through the lens of personal memory and experience. Instead of producing stable and homogeneous descriptions of images, this conceptual framework positions individual memory as a critical component of human-centered visual interpretation. When AI-recognized elements are associated with fragments of personal memory, visuals are reinterpreted and reimagined, transcending simple captions. What the AI sees is no longer separate from what individuals remember.

How I Perceive It is implemented using multiple state-of-the-art vision-language and generative models. Applied in an audiovisual art installation, these reinterpretations are expressed through memory-conditioned imagery, animated collages, and synthesized sound. Recognized elements, retrieved memories, and AI-reimagined visuals are overlaid and accompanied by glitch effects that reflect the instability and incompleteness of human recall. Narration in the artist’s cloned voice further blurs the boundary between internal memory and external mediation, forming a continuous yet unstable sense of self. From drawings to everyday visuals, a doodle and a photograph serve as instances through which images are reinterpreted into personally meaningful encounters.

More philosophical and literary background, as well as technical details and future applications, can be found in the proceedings of ACM SIGGRAPH Asia 2025 (How I Perceive It: Human Memory-Augmented Analogical Reasoning for Machine Visual Interpretation, Article No. 12, pp. 1–8).