Rethinking Inductive Biases for Surface Normal Estimation

Despite the growing demand for accurate <PRE_TAG>surface normal estimation</POST_TAG> models, existing methods use general-purpose <PRE_TAG>dense prediction models</POST_TAG>, adopting the same <PRE_TAG>inductive biases</POST_TAG> as other tasks. In this paper, we discuss the <PRE_TAG>inductive biases</POST_TAG> needed for <PRE_TAG>surface normal estimation</POST_TAG> and propose to (1) utilize the per-pixel ray direction and (2) encode the relationship between neighboring surface normals by learning their <PRE_TAG>relative rotation</POST_TAG>. The proposed method can generate crisp - yet, piecewise smooth - predictions for challenging in-the-wild images of arbitrary resolution and aspect ratio. Compared to a recent ViT-based state-of-the-art model, our method shows a stronger generalization ability, despite being trained on an orders of magnitude smaller dataset. The code is available at https://github.com/baegwangbin/DSINE.