Questions before
claims.
INsVI makes its research agenda visible while the work is underway. These inquiries guide experiments, dataset design, critical writing, and interdisciplinary collaboration.
What is lost when images become captions?
Captioned datasets preserve what can be named, but often discard composition, atmosphere, material process, visual tension, ambiguity, and culturally specific ways of seeing.
Beyond labels
What representations can retain relationships that ordinary labels flatten?
Evaluation
How can visual understanding be evaluated without turning every answer back into language?
Context
What provenance and cultural information must remain attached to an image?
Can a system reason through visual form?
We distinguish generating a plausible image from understanding spatial structure, physical relations, compositional decisions, and changes across time.
Space
Reasoning about depth, occlusion, scale, viewpoint, orientation, and navigation.
Composition
Recognizing hierarchy, rhythm, balance, contrast, and deliberate visual organization.
Time
Understanding persistence, transformation, sequence, and causality across visual observations.
Open research, documented carefully.
These are active research questions—not announcements of completed studies. INsVI will distinguish clearly among hypotheses, experiments, working papers, datasets, and peer-reviewed outcomes.