Vision models trained on chart images of time series often latch onto rendering style rather than temporal class structure, an effect the authors call visual encoding hijacking.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
VEIL: How Visual Encoding Hijacking Induces Bias In Vision Models
Vision models trained on chart images of time series often latch onto rendering style rather than temporal class structure, an effect the authors call visual encoding hijacking.