Learning Causal Representation for Training Cross-Domain Pose Estimator via Generative Interventions

Xi-heng, Zhang; Wong, Yongkang; Wu, Xiaofei; Lu, Juwei; Kankanhalli, Mohan; Li, Xiangdong; Wang, Geng

doi:10.1109/iccv48922.2021.01108

Cited by 18 publications

(7 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our method corroborates with results from previous works [28,29], which assert the existence of a problem regarding out-of-distribution unseen data and the possibility of using domain adaptation as a way to increase robustness in these scenarios. Furthermore, our method expands on existing techniques by not only tackling the challenge of domain adaptation but also by addressing the intricacies of pose misrepresentation.…”

Section: Discussionsupporting

confidence: 89%

“…This operated in such a way that minimizing the uncertainty for the unsupervised real dataset alongside a supervised synthetic dataset allows for the cross-dataset pose adaptation. Zhang et al [29], on the other hand, proposed a method for learning causal representations in order to generate out-of-distribution features that can properly generalize to unseen domains. Some works tried to solve the problem of pose misrepresentation, which is also found in the literature regarding cross-dataset evaluation.…”

Section: Cross-domain 3d Human Pose Estimationmentioning

confidence: 99%

“…Cross-dataset evaluation in 3D human pose estimation shows a significant challenge due to the inherent misalignment of target distributions, especially when synthetic data are involved. The scarcity of literature addressing this specific scenario has motivated only a few authors to explore the evaluation protocol for assessing cross-domain generalization when synthetic data are involved [28,29].…”

Section: Evaluation Protocolmentioning

confidence: 99%

See 2 more Smart Citations

DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose Estimation

Manesco,

Berretti,

Marana

2023

Sensors

View full text Add to dashboard Cite

Human pose estimation is an important Computer Vision problem, whose goal is to estimate the human body through joints. Currently, methods that employ deep learning techniques excel in the task of 2D human pose estimation. However, the use of 3D poses can bring more accurate and robust results. Since 3D pose labels can only be acquired in restricted scenarios, fully convolutional methods tend to perform poorly on the task. One strategy to solve this problem is to use 2D pose estimators, to estimate 3D poses in two steps using 2D pose inputs. Due to database acquisition constraints, the performance improvement of this strategy can only be observed in controlled environments, therefore domain adaptation techniques can be used to increase the generalization capability of the system by inserting information from synthetic domains. In this work, we propose a novel method called Domain Unified approach, aimed at solving pose misalignment problems on a cross-dataset scenario, through a combination of three modules on top of the pose estimator: pose converter, uncertainty estimator, and domain classifier. Our method led to a 44.1mm (29.24%) error reduction, when training with the SURREAL synthetic dataset and evaluating with Human3.6M over a no-adaption scenario, achieving state-of-the-art performance.

show abstract

Section: Discussionsupporting

confidence: 89%

Section: Cross-domain 3d Human Pose Estimationmentioning

confidence: 99%

Section: Evaluation Protocolmentioning

confidence: 99%

See 1 more Smart Citation

DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose Estimation

Manesco,

Berretti,

Marana

2023

Sensors

View full text Add to dashboard Cite

show abstract

“…Rather than maximising accuracy on a hold-out test set, one may resort to CDL to also achieve high accuracy on a subset that is not representative of the training data. Examples such as these include: domain adaptation [24,85], transfer learning [86,87], interpretability [88][89][90][91], or general robustness [92][93][94][95].…”

Section: Why Use CDL For Supervised Learning?mentioning

confidence: 99%

Causal Deep Learning

Berrevoets¹,

Kacprzyk²,

Qian³

et al. 2023

Preprint

View full text Add to dashboard Cite

Causality has the potential to truly transform the way we solve a large number of realworld problems. Yet, so far, its potential remains largely unlocked since most work so far requires strict assumptions which do not hold true in practice. To address this challenge and make progress in solving real-world problems, we propose a new way of thinking about causality -we call this causal deep learning. The framework which we propose for causal deep learning spans three dimensions: (1) a structural dimension, which allows incomplete causal knowledge rather than assuming either full or no causal knowledge; (2) a parametric dimension, which encompasses parametric forms which are typically ignored; and finally, (3) a temporal dimension, which explicitly allows for situations which capture exposure times or temporal structure. Together, these dimensions allow us to make progress on a variety of real-world problems by leveraging (sometimes incomplete) causal knowledge and/or combining diverse causal deep learning methods. This new framework also enables researchers to compare systematically across existing works as well as identify promising research areas which can lead to real-world impact.

show abstract

“…Cross-Domain Pose Estimator: Zhang et al [287] propose using causal representation learning to improve cross-domain 3D pose estimation tasks. Specifically, they train a counterfactual feature generator that takes domains and contents as input.…”

Section: Counterfactual Data Augmentationmentioning

confidence: 99%

Causal Machine Learning: A Survey and Open Problems

Jean¹,

Lynch²,

Liu³

et al. 2022

Preprint

View full text Add to dashboard Cite

Causal Machine Learning (CausalML) is an umbrella term for machine learning methods that formalize the data-generation process as a structural causal model (SCM). This allows one to reason about the effects of changes to this process (i.e., interventions) and what would have happened in hindsight (i.e., counterfactuals). We categorize work in CausalML into five groups according to the problems they tackle: (1) causal supervised learning, (2) causal generative modeling, (3) causal explanations, (4) causal fairness, (5) causal reinforcement learning. For each category, we systematically compare its methods and point out open problems. Further, we review modality-specific applications in computer vision, natural language processing, and graph representation learning. Finally, we provide an overview of causal benchmarks and a critical discussion of the state of this nascent field, including recommendations for future work.

show abstract

Learning Causal Representation for Training Cross-Domain Pose Estimator via Generative Interventions

Cited by 18 publications

References 44 publications

DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose Estimation

DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose Estimation

Causal Deep Learning

Causal Machine Learning: A Survey and Open Problems

Contact Info

Product

Resources

About