SpuriVerse, a benchmark of 124 real-world spurious correlation types, shows LVLMs fail on such cases and that diverse synthetic training improves robustness to unseen spurious correlations.
a model with good accuracy on SpuriVerse can be free from all 18 potential spurious correlation attacks,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Escaping the SpuriVerse: Can Large Vision-Language Models Generalize Beyond Seen Spurious Correlations?
SpuriVerse, a benchmark of 124 real-world spurious correlation types, shows LVLMs fail on such cases and that diverse synthetic training improves robustness to unseen spurious correlations.