DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose Estimation
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. Sinc...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-08-01
|
| Series: | Sensors |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1424-8220/23/17/7312 |
| _version_ | 1827727807114903552 |
|---|---|
| author | João Renato Ribeiro Manesco Stefano Berretti Aparecido Nilceu Marana |
| author_facet | João Renato Ribeiro Manesco Stefano Berretti Aparecido Nilceu Marana |
| author_sort | João Renato Ribeiro Manesco |
| collection | DOAJ |
| description | 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. |
| first_indexed | 2024-03-10T23:13:14Z |
| format | Article |
| id | doaj.art-43ca4a623bf14ecd8fc6d7747fe2c35c |
| institution | Directory Open Access Journal |
| issn | 1424-8220 |
| language | English |
| last_indexed | 2024-03-10T23:13:14Z |
| publishDate | 2023-08-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Sensors |
| spelling | doaj.art-43ca4a623bf14ecd8fc6d7747fe2c35c2023-11-19T08:48:11ZengMDPI AGSensors1424-82202023-08-012317731210.3390/s23177312DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose EstimationJoão Renato Ribeiro Manesco0Stefano Berretti1Aparecido Nilceu Marana2Faculty of Sciences, UNESP—São Paulo State University, Bauru 17033-360, SP, BrazilMedia Integration and Communication Center (MICC), University of Florence, 50134 Florence, ItalyFaculty of Sciences, UNESP—São Paulo State University, Bauru 17033-360, SP, BrazilHuman 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.https://www.mdpi.com/1424-8220/23/17/73123D human pose estimationdomain adaptationadversarial neural networks |
| spellingShingle | João Renato Ribeiro Manesco Stefano Berretti Aparecido Nilceu Marana DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose Estimation Sensors 3D human pose estimation domain adaptation adversarial neural networks |
| title | DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose Estimation |
| title_full | DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose Estimation |
| title_fullStr | DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose Estimation |
| title_full_unstemmed | DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose Estimation |
| title_short | DUA: A Domain-Unified Approach for Cross-Dataset 3D Human Pose Estimation |
| title_sort | dua a domain unified approach for cross dataset 3d human pose estimation |
| topic | 3D human pose estimation domain adaptation adversarial neural networks |
| url | https://www.mdpi.com/1424-8220/23/17/7312 |
| work_keys_str_mv | AT joaorenatoribeiromanesco duaadomainunifiedapproachforcrossdataset3dhumanposeestimation AT stefanoberretti duaadomainunifiedapproachforcrossdataset3dhumanposeestimation AT aparecidonilceumarana duaadomainunifiedapproachforcrossdataset3dhumanposeestimation |