Two papers from NTT Laboratories have been accepted for publication for WACV2026

Two papers authored by NTT laboratories have been accepted at WACV2026 (Winter Conference on Applications of Computer Vision) 2026, to be held in Tucson, Arizona, USA, from March 6 to March 10, 2026. WACV is known as one of the most prestigious international conferences in the field of computer vision.

Abbreviated names of the laboratories:
HI: NTT Human Informatics Laboratories
CD: NTT Computer and Data Science Laboratories
CS: NTT Communication Science Laboratories

■IPCD: Intrinsic Point-Cloud Decomposition

Shogo Sato (HI), Takuhiro Kaneko (CS), Shoichiro Takeda (HI), Tomoyasu Shimada (HI), Kazuhiko Murasaki (HI), Taiga Yoshida (HI), Ryuichi Tanida (HI), Akisato Kimura (CS)

This study introduces a new task, Intrinsic Point-Cloud Decomposition, which separates colored point clouds into albedo (reflectance) and shading (illumination) components. To address the challenge of estimating lighting in non-grid point cloud structures, the proposed IPCD-Net incorporates point-wise feature aggregation and PLD, a multi-view–projection–based luminance distribution estimator. Evaluated on a newly constructed dataset, the method achieves more accurate shadow removal and improved color fidelity compared to previous approaches. The paper also demonstrates practical applications such as relighting and point-cloud editing.

■Distribution Highlighted Reference-Based Label Distribution Learning for Facial Age Estimation

Satoshi Suzuki (HI), Shin'ya Yamaguchi (CD), Shoichiro Takeda (HI), Takuhiro Kaneko (CS), Shota Orihashi (HI), Ryo Masumura (HI)

In facial age estimation, a unique challenge is that faces of the same individual across similar ages are often difficult to distinguish. To address this issue, approaches that predict a label distribution, i.e., the probability that an image belongs to each age, instead of a single age label, have been widely adopted. However, the heuristic constraints used in existing methods make it difficult to predict label distributions that fully reflect the characteristics of individual input images. In this work, we propose a novel age estimation method that introduces input-dependent constraints to address this problem.