Best Paper Honorable Mention Award at the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026 – Subtle Visual Computing Workshop

A new study by members of The Computer Vision and Medical Lab, VinUni-Illinois Smart Health Center (VISHC), VinUniversity has been honored with the Best Paper Honorable Mention Award at The 2nd Workshop on Subtle Visual Computing (SVC), held in conjunction with CVPR 2026 (Conference on Computer Vision and Pattern Recognition) in Denver, Colorado. SVC Workshop is an official CVPR 2026 workshop focused on subtle visual computing—the study of detecting and interpreting subtle visual cues in images and videos, with applications in healthcare, biometric security, industrial inspection, and affective computing.

The study, titled “BTS-rPPG: Orthogonal Butterfly Temporal Shifting for Remote Photoplethysmography,” by Ba-Thinh Nguyen et al., focuses on remote photoplethysmography, or rPPG, a vision-based technique for estimating physiological signals such as heart rate from facial videos without requiring physical contact sensors.

Figure 1: Best Paper Honorable Mention Award Certificate

The research addresses a key challenge in rPPG: modeling the temporal dynamics of subtle pulse-related facial appearance changes. Existing temporal shifting and convolution-based methods often exchange information mainly between neighboring frames, resulting in predominantly local temporal modeling and limited temporal receptive fields. In addition, direct feature transfer across frames may introduce redundant information, since facial video frames are often highly correlated over time. 

Figure 2: Overview of the proposed BTS-rPPG. It enables AI to measure physiological signals from a simple facial video by focusing on subtle temporal changes and efficiently combining information across time. 

To tackle these limitations, the team proposed BTS-rPPG, a temporal modeling framework based on Butterfly Temporal Shifting. Inspired by the butterfly communication pattern in the Fast Fourier Transform, BTS establishes structured frame interactions through an XOR-based pairing schedule, progressively expanding temporal communication from short-range to long-range dependencies. The framework further introduces Orthogonal Feature Transfer, which filters the source feature with respect to the target context before shifting, allowing only complementary information to be transferred across frames.

Extensive experiments on multiple rPPG benchmark datasets demonstrate that BTS-rPPG consistently improves heart-rate estimation performance under both intra-dataset and cross-dataset settings. The results show that the proposed framework can better capture long-range physiological dynamics while reducing redundant temporal communication.

This achievement highlights VinUniversity’s growing research strength in computer vision, biomedical AI, and remote physiological sensing, and marks an important recognition for the team at an internationally recognized CVPR workshop.

Full paper can be found here: https://shorturl.at/LVdoO