Proposes LDCBM, a lightweight extension of concept bottleneck models that disentangles features into concepts without region annotations to mitigate background bias and improve accuracy and interpretability.
VLG-CBM: Training Concept Bottleneck Models with Vision-Language Guidance,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Mitigating Spurious Background Bias in Multimedia Recognition with Disentangled Concept Bottlenecks
Proposes LDCBM, a lightweight extension of concept bottleneck models that disentangles features into concepts without region annotations to mitigate background bias and improve accuracy and interpretability.