Relative Pose Consistency for Semi-Supervised Head Pose Estimation

Kuhnke, Felix; Ihler, Sontje; Östermann, Jörn

doi:10.1109/fg52635.2021.9666992

Cited by 1 publication

(13 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Lastly, it is a common approach to use synthetic face datasets from 3D models for HPE [11], [12], [13], [14], [15], [16], [20]. This has the advantage of learning from a high amount of diverse images with perfect labels.…”

Section: Head Pose Estimationmentioning

confidence: 99%

“…These related works train on synthetic or mixed datasets and evaluate on real-world datasets, however, most of them do not explore any domain adaptation or semi-supervised techniques to overcome the domain gap. This was only explored in our previous works [16], [20].…”

Section: Head Pose Estimationmentioning

confidence: 99%

“…In this work, an extended version of [20], our goal is to improve the performance of HPE on real-world data using only labels from a synthetic dataset in combination with an unlabeled real-world dataset. In [16] an adversarial training approach based on domain adversarial neural networks [21] is used to force the extraction of domain-invariant features.…”

Section: Introductionmentioning

confidence: 99%

“…pose is unknown for our target-domain data. In [20] and this extended work, we therefore propose to take advantage of relative pose.…”

Section: Introductionmentioning

confidence: 99%

“…We present an improved training framework compared to [20] • We uncover effects of preprocessing on performance during reproduction of related work.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Domain Adaptation for Head Pose Estimation Using Relative Pose Consistency

Kuhnke

Östermann

2023

IEEE Trans. Biom. Behav. Identity Sci.

View full text Add to dashboard Cite

Head pose estimation plays a vital role in biometric systems related to facial and human behavior analysis. Typically, neural networks are trained on head pose datasets. Unfortunately, manual or sensor-based annotation of head pose is impractical. A solution is synthetic training data generated from 3D face models, which can provide an infinite number of perfect labels. However, computer generated images only provide an approximation of real-world images, leading to a performance gap between training and application domain. Therefore, there is a need for strategies that allow simultaneous learning on labeled synthetic data and unlabeled real-world data to overcome the domain gap. In this work we propose relative pose consistency, a semi-supervised learning strategy for head pose estimation based on consistency regularization. Consistency regularization enforces consistent network predictions under random image augmentations, including pose-preserving and pose-altering augmentations. We propose a strategy to exploit the relative pose introduced by pose-altering augmentations between augmented image pairs, to allow the network to benefit from relative pose labels during training on unlabeled data. We evaluate our approach in a domain-adaptation scenario and in a commonly used cross-dataset scenario. Furthermore, we reproduce related works to enforce consistent evaluation protocols and show that for both scenarios we outperform SOTA.

show abstract

Section: Head Pose Estimationmentioning

confidence: 99%

Section: Head Pose Estimationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

“…pose is unknown for our target-domain data. In [20] and this extended work, we therefore propose to take advantage of relative pose.…”

Section: Introductionmentioning

confidence: 99%

“…We present an improved training framework compared to [20] • We uncover effects of preprocessing on performance during reproduction of related work.…”

mentioning

confidence: 99%

See 3 more Smart Citations