Lune

CVPR2023Top-tier venue

Recurrence without Recurrence: Stable Video Landmark Detection with Deep Equilibrium Models

Paul Micaelli, Arash Vahdat, Hongxu Yin, Jan Kautz, Pavlo Molchanov

Abstract

Cascaded computation, whereby predictions are recurrently refined over several stages, has been a persistent theme throughout the development of landmark detection models. In this work, we show that the recently proposed Deep Equilibrium Model (DEQ) can be naturally adapted to this form of computation. Our Landmark DEQ (LDEQ) achieves state-of-the-art performance on the challenging WFLW facial landmark dataset, reaching 3.92 NME with fewer parameters and a training memory cost of O(1) in the number of recurrent modules. Furthermore, we show that DEQs are particularly suited for landmark detection in videos. In this setting, it is typical to train on still images due to the lack of labelled videos. This can lead to a "flickering" effect at inference time on video, whereby a model can rapidly oscillate between different plausible solutions across consecutive frames. By rephrasing DEQs as a constrained optimization, we emulate recurrence at inference time, despite not having access to temporal data at training time. This Recurrence without Recurrence (RwR) paradigm helps in reducing landmark flicker, which we demonstrate by introducing a new metric, normalized mean flicker (NMF), and contributing a new facial landmark video dataset (WFLW-V) targeting landmark uncertainty. On the WFLW-V hard subset made up of 500 videos, our LDEQ with RwR improves the NME and NMF by 10 and 13% respectively, compared to the strongest previously published model using a hand-tuned conventional filter.

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.

9 top-tier papers have cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Adaptive Risk Minimization: Learning to Adapt to Domain Shift cites Tent: Fully Test-Time Adaptation by Entropy Minimization
  • Adaptive Risk Minimization: Learning to Adapt to Domain Shift cites Test-Time Training with Self-Supervision for Generalization under Distribution Shifts
  • One-Step Diffusion Distillation via Deep Equilibrium Models cites Multiscale Deep Equilibrium Models
  • Heatmap Regression without Soft-Argmax for Facial Landmark Detection cites Adaptive Wing Loss for Robust Face Alignment via Heatmap Regression
  • Exploiting Connections between Lipschitz Structures for Certifiably Robust Deep Equilibrium Models cites Multiscale Deep Equilibrium Models
  • Tent: Fully Test-Time Adaptation by Entropy Minimization cites Multiscale Deep Equilibrium Models
  • Deep Equilibrium Models are Almost Equivalent to Not-so-deep Explicit Models for High-dimensional Gaussian Mixtures cites Multiscale Deep Equilibrium Models
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers9

  1. One-Step Diffusion Distillation via Deep Equilibrium ModelsZhengyang Geng, Ashwini Pokle, J. Zico KolterNeurIPS 2023 · 83 citations
  2. FaceXFormer: A Unified Transformer for Facial AnalysisKartik Narayan, Vibashan VS, Rama Chellappa, Vishal M. PatelICCV 2025 · 16 citations
  3. Exploiting Connections between Lipschitz Structures for Certifiably Robust Deep Equilibrium ModelsAaron J. Havens, Alexandre Araujo, Siddharth Garg, Farshad Khorrami et al.NeurIPS 2023 · 15 citations
  4. Deep Equilibrium Models are Almost Equivalent to Not-so-deep Explicit Models for High-dimensional Gaussian MixturesZenan Ling, Longbo Li, Zhanbo Feng, Yixuan Zhang et al.ICML 2024 · 6 citations
  5. Heatmap Regression without Soft-Argmax for Facial Landmark DetectionChiao-An Yang, Raymond A. YehICCV 2025 · 3 citations

Builds on21

  1. Tent: Fully Test-Time Adaptation by Entropy MinimizationDequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno A. Olshausen et al.ICLR 2021 · 1,731 citations
  2. Test-Time Training with Self-Supervision for Generalization under Distribution ShiftsYu Sun, Xiaolong Wang, Zhuang Liu, John Miller et al.ICML 2020 · 1,220 citations
  3. Adaptive Wing Loss for Robust Face Alignment via Heatmap RegressionXinyao Wang, Liefeng Bo, Fuxin LiICCV 2019 · 293 citations
  4. Adaptive Risk Minimization: Learning to Adapt to Domain ShiftMarvin Zhang, Henrik Marklund, Nikita Dhawan, Abhishek Gupta et al.NeurIPS 2021 · 284 citations
  5. Multiscale Deep Equilibrium ModelsShaojie Bai, Vladlen Koltun, J. Zico KolterNeurIPS 2020 · 272 citations

Related papers

  1. FAB: A Robust Facial Landmark Detection Framework for Motion-Blurred VideosKeqiang Sun, Wayne Wu, Tinghao Liu, Shuo Yang et al.ICCV 2019 · 32 citations
  2. Laplace Landmark LocalizationJoseph P. Robinson, Yuncheng Li, Ning Zhang, Yun Fu et al.ICCV 2019 · 49 citations
  3. STAR Loss: Reducing Semantic Ambiguity in Facial Landmark DetectionZhenglin Zhou, Huaxia Li, Hong Liu, Nanyang Wang et al.CVPR 2023
  4. Exploiting Self-Supervised and Semi-Supervised Learning for Facial Landmark Tracking with Unlabeled DataShi Yin, Shangfei Wang, Xiaoping Chen, Enhong ChenACM MM 2020 · 7 citations
  5. Progressive Guessing to Fixed Point: Rethinking Human Motion Prediction with Deep Equilibrium ModelsDong Wei, Huaijiang Sun, Fan Liu, Yuhui ZhengCVPR 2026 · 1 citation
Dusk over the sea between two cliffs drawn in fine vertical lines