Subtle visual signals, though often imperceptible to the human eye, contain subtle yet crucial information that can reveal hidden patterns within visual data. By applying advanced computer vision and representation learning techniques, we can unlock the potential of these signals to better understand and interpret complex environments. This ability to detect and analyze subtle signals has profound implications across various fields (see Figure below), e.g., (1) from biometric and multimedia security, where detecting high-fidelity physical spoof and digital manipulation can enhance the trustworthiness and reliablity of real-world visual systems; (2) from medicine, where early identification of minute anomalies in medical imaging can lead to life-saving interventions, (3) from industry, where spotting micro-defects in production lines can prevent costly failures, (4) from affective computing, where understanding micro-expression, micro-gesture, and hidden physiological signals under human interaction scenarios can benefit the deception detection. In an era overwhelmed by information, the capacity to detect and decode these subtle visual signals offers a novel and powerful approach to anticipating trends, identifying emerging threats, and discovering new opportunities. These signals, often ignored or overlooked, may hold key insights into future developments across different societal contexts.

Although recent advances in subtle visual computing have demonstrated significant potential, several challenges persist in terms of effectiveness, robustness, and generalization. Specifically, these challenges include:

•  Limited Representation of Subtle Visual Signals: Existing visual models rely on specific operators in the spatial and temporal domains to perceive subtle clues. However, subtle signals in the spatio-temporal domain are easily affected by visual quality degradation and harsh environmental interference or even drowned out, resulting in poor model representation and generalization capacity. Faced with extremely subtle fine-grained contextual information in the spatio-temporal domain in complex open-world environments, exploring next-generation foundation models with advanced operators and modules is a key issue for subtle visual computing.

• Limited Performance in Multi-task and Multimodal Scenarios: Existing visual models focus on subtle signal representation for specific tasks and modalities, such as face attack detection, remote vital sign measurement, and camouflaged object segmentation, within fixed modality settings such as visible light, near-infrared, and depth. However, different types of related tasks can provide useful complementary information. Thus, cross-task and cross-modal complementary information is often overlooked. It is necessary to explore a unified framework to enhance the model's ability to express subtle visual signals in multi-task and cross-modal scenarios.

This workshop seeks to develop innovative representation learning models specifically designed to capture and interpret subtle visual signals. By doing so, it will provide new ways of perceiving and acting on visual information, empowering decision-making in fields such as security, healthcare, industrial processes, and affective computing. Ultimately, we hope this special issue aspires to demonstrate how hidden visual cues, when properly decoded, can offer critical foresight and actionable insights in an increasingly complex and interconnected world.

Organizing Committee

Invited Keynote Speakers

Challenge and Data Chairs

2nd SVC Workshop Details

Important Dates:

• Submission Start:  1 February, 2026

• Submission Deadline:  20 March, 2026

• Acceptance Notification:  1 April, 2026

• Camera-Ready Paper:  10 April, 2026

Topics of Interest:

The proposed workshop encourages various subtle visual topics, including but not limited to:

• Theoretical analysis of robustness, generalization, and interpretability in subtle visual computing;

• Bio-inspired and human visual system inspired subtle computing

• Subtle visual signal magnification;

• Image and video based camouflaged object detection;

• Subtle visual anomaly detection in medicine, industry, and biometric systems;

• Subtle multimedia manipulation detection and localization; 

• Subtle change detection for remote sensing;

• Subtle human behavior understanding (e.g., micro-expression & micro-gesture analysis, deception detection);

• Video-based subtle physiological signal measurement;

• New synthesis models for subtle visual content generation;

• Innovative learning strategies for multi-modal subtle visual representation; 

• Lightweight and general backbone designs for subtle visual computing; 

• Large-scale datasets and benchmarks specific to task-specific or unified subtle visual computing;

• Survey or technical review of recent advances in subtle visual computing

Submission Guidelines:

• Paper presented at SVC-CVPR26 will be published as an official CVPR Workshop proceedings, and follow the same guidelines as the main conference of CVPR 2026.

• Submit your papers at: https://openreview.net/group?id=thecvf.com/CVPR/2026/Workshop/SVC

• The LaTex/Word templates for the paper submission can be found in Paper Submission.

• Page limit: A paper can be up to 8 pages, including figures and tables, plus an unlimited number of additional pages for references only. 

• Papers will be double-blind peer-reviewed by at least two reviewers. Please remove author names, affiliations, email addresses, etc. from the paper. Remove personal acknowledgments.

• The best workshop paper of SVC will be recommended to be extended to the journal Machine Intelligence Research or Journal of Image and Graphics. 

Please register and participate in the challenge at  https://www.codabench.org/competitions/12678.

Overview:

Multimodal deception detection (MMDD) [1,2,3] is a typical subtle visual computing task, aiming to detect imperceptible and deceptive clues from audio-visual scenarios. The Multimodal Deception Detection Competition aims to bring together researchers and developers to advance the field of multimodal learning by detecting deception through the integration of multiple modalities such as audio, video, and text. The competition encourages innovation in building robust AI models that can accurately identify deceptive behaviors by leveraging various features from these modalities.

Please register and participate in the challenge at  https://www.codabench.org/competitions/12857.

Overview:

The Domain Generalized Remote Physiological Measurement (PhysDG) [4] is designed to advance research on robust, domain-invariant rPPG algorithms. Participants are encouraged to explore strategies that enhance cross-domain generalization through improved feature learning, domain adaptation, or meta-learning approaches. As an optional extension, participants may also predict additional physiological parameters such as respiratory rate, blood pressure, or blood glucose, although multi-parameter prediction is not mandatory.

Important Dates:

• Competition Launch & Data Release:  1 February, 2026

• Registration Deadline:  1 March, 2026 

  • Stage 1:   1 February, 2026 - 1 March, 2026

  • Stage 2:   1 March, 2026 - 15 March, 2026

• Top-3 Winners (Must submit workshop papers with registration) Announcement:  20 March, 2026

• The winners will be encouraged to extend the workshop paper to the journal Machine Intelligence Research. 

Reference:

[1] Xun Lin, Xiaobao Guo, Taorui Wang, Yingjie Ma, Jiajian Huang, Jiayu Zhang, Junzhe Cao, Zitong Yu. SVC 2025: the First Multimodal Deception Detection Challenge, ACM MM Workshop of SVC, 2025

[2] Xiaobao Guo, Nithish Muthuchamy Selvaraj, Zitong Yu, Adams Kong, Bingquan Shen, Alex Kot. Audio-Visual Deception Detection: DOLOS Dataset and Parameter-Efficient Crossmodal Learning, ICCV 2023

[3] Xiaobao Guo, Zitong Yu, Nithish Muthuchamy Selvaraj, Bingquan Shen, Adams Wai-Kin Kong, Alex C Kot. Benchmarking Cross-Domain Audio-Visual Deception Detection, arXiv 2024

[4] Zitong Yu, Yuming Shen, Jingang Shi, Hengshuang Zhao, Philip Torr, Guoying Zhao. PhysFormer: Facial Video-based Physiological Measurement with Temporal Difference Transformer, CVPR 2022

1st Workshop & Challenge on  Subtle Visual Computing (SVC) @ ACM MM 2025 can be found at https://sites.google.com/view/svc-mm25