CURVE: Learning Causality-Inspired Invariant Representations for Robust Scene Understanding via Uncertainty-Guided Regularization
This work addresses the robustness of scene understanding models for computer vision applications, representing an incremental improvement through a novel regularization approach.
The paper tackles the problem of scene graph models overfitting to spurious correlations, which hinders out-of-distribution generalization, by proposing CURVE, a causality-inspired framework that integrates variational uncertainty modeling with uncertainty-guided regularization to suppress environment-specific relations. The result is improved zero-shot transfer and low-data sim-to-real adaptation, with the model learning domain-stable sparse topologies and providing reliable uncertainty estimates for risk prediction under distribution shifts.
Scene graphs provide structured abstractions for scene understanding, yet they often overfit to spurious correlations, severely hindering out-of-distribution generalization. To address this limitation, we propose CURVE, a causality-inspired framework that integrates variational uncertainty modeling with uncertainty-guided structural regularization to suppress high-variance, environment-specific relations. Specifically, we apply prototype-conditioned debiasing to disentangle invariant interaction dynamics from environment-dependent variations, promoting a sparse and domain-stable topology. Empirically, we evaluate CURVE in zero-shot transfer and low-data sim-to-real adaptation, verifying its ability to learn domain-stable sparse topologies and provide reliable uncertainty estimates to support risk prediction under distribution shifts.