Lune

CVPR Findings2026

Weakly Supervised Micro-Expression Spotting based on Boundary Refinement Mechanism and Cross-subject Learning Representation

Zhihua Xie, Haolin Chang, Guohua Miao, Jianing Chen

Abstract

Weakly supervised learning paradigms present promising potential for micro-expression (ME) analysis. However, existing weakly supervised micro-expression spotting (MES) methods still suffer from temporal ambiguity in boundary spotting and semantic inconsistency in cross-subject feature representation. The temporal ambiguity in boundary spotting arises from the constraint that neural networks can only learn from video-level labels under weakly supervised settings, substantially increasing the difficulty of precise temporal spotting. The ambiguity in cross-subject feature representation primarily manifests as intra-class heterogeneity and inter-class homogeneity within cross-subject feature spaces. To address these issues, this paper proposes a Weakly Supervised MES Model based on the boundary refinement mechanism and cross-subject learning representation, named WS-MES. Specifically, the boundary refinement mechanism is leveraged to facilitate precise boundary location from coarse-grained pseudo-labeled candidate proposals. In response to cross-subject representation ambiguity, the WS-MES establishes the multi-level consistency constraint. On the one hand, the semantic consistency constraint based on optimal transport is integrated to enhance cross-subject intra-class semantic compactness and suppress semantic drift. On the other hand, the discriminative consistency constraint using prototype contrastive learning is incorporated to aggregate crosssubject inter-class discriminability. Extensive experimental results show that WS-MES substantially outperforms mainstream weakly-supervised MES approaches.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

Builds onCited byLeft to right by year; the more cited, the higher.

Builds on10

  1. Self-labelling via simultaneous clustering and representation learningYuki Markus Asano, Christian Rupprecht, Andrea VedaldiICLR 2020 · 873 citations
  2. Learning Action Completeness from Points for Weakly-supervised Temporal Action LocalizationPilhyeon Lee, Hyeran ByunICCV 2021 · 81 citations
  3. Multi-Level Optimal Transport for Universal Cross-Tokenizer Knowledge Distillation on Language ModelsXiao Cui, Mo Zhu, Yulei Qin, Liang Xie et al.AAAI 2025 · 31 citations
  4. Weakly-Supervised Temporal Action Localization by Inferring Salient Snippet-FeatureWulian Yun, Mengshi Qi, Chuanming Wang, Huadong MaAAAI 2024 · 29 citations
  5. Actionness Inconsistency-Guided Contrastive Learning for Weakly-Supervised Temporal Action LocalizationZhilin Li, Zilei Wang, Qinying LiuAAAI 2023 · 12 citations

Related papers

  1. Face-Guided Sentiment Boundary Enhancement for Weakly-Supervised Temporal Sentiment LocalizationCailing Han, Zhangbin Li, Jinxing Zhou, Wei Qian et al.CVPR 2026 · 1 citation
  2. Exploring Denoised Cross-video Contrast for Weakly-supervised Temporal Action LocalizationJingjing Li, Tianyu Yang, Wei Ji, Jue Wang et al.CVPR 2022 · 57 citations
  3. Weakly-Supervised Audio-Visual Video Parsing with Prototype-Based Pseudo-LabelingKranthi Kumar Rachavarapu, Kalyan Ramakrishnan, A. N. RajagopalanCVPR 2024 · 3 citations
  4. Prompt-Enhanced Multiple Instance Learning for Weakly Supervised Video Anomaly DetectionJunxi Chen, Liang Li, Li Su, Zheng-Jun Zha et al.CVPR 2024
  5. Weakly-Supervised Temporal Action Alignment Driven by Unbalanced Spectral Fused Gromov-Wasserstein DistanceDixin Luo, Yutong Wang, Angxiao Yue, Hongteng XuACM MM 2022 · 8 citations
Dusk over the sea between two cliffs drawn in fine vertical lines