Constrained visual reconstruction
What information is lost first when natural images are reconstructed from a finite alphabet of glyphs, and which local image statistics best preserve perceived structure?
ACTIVE QUESTIONS / METHODS / VALIDATION
Research
A structured index of research directions. Entries here are intentionally scoped as questions and methods rather than as polished project claims.
Inspectable multimodal inference
How can multimodal pipelines expose intermediate decisions well enough to validate screening, classification, extraction, and routing independently?
Low-resolution embodied perception
Which visual representations remain sufficient for target recognition and obstacle-aware navigation when sensing is deliberately constrained?
Interfaces for model uncertainty
What interface structures help technically sophisticated users distinguish model confidence, missing evidence, and unresolved ambiguity without collapsing them into one score?
M-01Photometric analysis
Luminance estimation, local contrast, illumination fields, shadow-like darkness.
M-02Gradient structure
Sobel/Canny families, local orientation, contour retention, boundary weighting.
M-03Validation
Gold labels, precision/recall/F1, confusion analysis, failure strata, ablation.
M-04Interactive inspection
Executable demos used to expose intermediate representations and failure cases.