Geometry-Biased Transformer for Robust Multi-View 3D Human Pose Reconstruction
(2024) 18th IEEE International Conference on Automatic Face and Gesture Recognition, FG 2024- Abstract
We address the challenges in estimating 3D human poses from multiple views under occlusion and with limited overlapping views. We approach multi-view, single-person 3D human pose reconstruction as a regression problem and propose a novel encoder-decoder Transformer architecture to estimate 3D poses from multi-view 2D pose sequences. The encoder refines 2D skeleton joints detected across different views and times, fusing multi-view and temporal information through global self-attention. We enhance the encoder by incorporating a geometry-biased attention mechanism, effectively leveraging geometric relationships between views. Additionally, we use detection scores provided by the 2D pose detector to further guide the encoder's attention... (More)
We address the challenges in estimating 3D human poses from multiple views under occlusion and with limited overlapping views. We approach multi-view, single-person 3D human pose reconstruction as a regression problem and propose a novel encoder-decoder Transformer architecture to estimate 3D poses from multi-view 2D pose sequences. The encoder refines 2D skeleton joints detected across different views and times, fusing multi-view and temporal information through global self-attention. We enhance the encoder by incorporating a geometry-biased attention mechanism, effectively leveraging geometric relationships between views. Additionally, we use detection scores provided by the 2D pose detector to further guide the encoder's attention based on the reliability of the 2D detections. The decoder subsequently regresses the 3D pose sequence from these refined tokens, using pre-defined queries for each joint. To enhance the generalization of our method to unseen scenes and improve resilience to missing joints, we implement strategies including scene centering, synthetic views, and token dropout. We conduct extensive experiments on three benchmark public datasets, Human3.6M, CMU Panoptic and Occlusion-Persons. Our results demonstrate the efficacy of our approach, particularly in occluded scenes and when few views are available, which are traditionally challenging scenarios for triangulation-based methods.
(Less)
- author
- Moliner, Olivier LU ; Huang, Sangxia and Astrom, Kalle LU
- organization
- publishing date
- 2024-07-11
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- host publication
- 2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition, FG 2024
- publisher
- IEEE - Institute of Electrical and Electronics Engineers Inc.
- conference name
- 18th IEEE International Conference on Automatic Face and Gesture Recognition, FG 2024
- conference location
- Istanbul, Turkey
- conference dates
- 2024-05-27 - 2024-05-31
- external identifiers
-
- scopus:85199438076
- ISBN
- 9798350394948
- DOI
- 10.1109/FG59268.2024.10581930
- language
- English
- LU publication?
- yes
- additional info
- Publisher Copyright: © 2024 IEEE.
- id
- 04c52fd8-f45a-4147-812f-4591b6b303e3
- date added to LUP
- 2024-08-14 14:54:17
- date last changed
- 2024-09-13 03:10:36
@inproceedings{04c52fd8-f45a-4147-812f-4591b6b303e3, abstract = {{<p>We address the challenges in estimating 3D human poses from multiple views under occlusion and with limited overlapping views. We approach multi-view, single-person 3D human pose reconstruction as a regression problem and propose a novel encoder-decoder Transformer architecture to estimate 3D poses from multi-view 2D pose sequences. The encoder refines 2D skeleton joints detected across different views and times, fusing multi-view and temporal information through global self-attention. We enhance the encoder by incorporating a geometry-biased attention mechanism, effectively leveraging geometric relationships between views. Additionally, we use detection scores provided by the 2D pose detector to further guide the encoder's attention based on the reliability of the 2D detections. The decoder subsequently regresses the 3D pose sequence from these refined tokens, using pre-defined queries for each joint. To enhance the generalization of our method to unseen scenes and improve resilience to missing joints, we implement strategies including scene centering, synthetic views, and token dropout. We conduct extensive experiments on three benchmark public datasets, Human3.6M, CMU Panoptic and Occlusion-Persons. Our results demonstrate the efficacy of our approach, particularly in occluded scenes and when few views are available, which are traditionally challenging scenarios for triangulation-based methods.</p>}}, author = {{Moliner, Olivier and Huang, Sangxia and Astrom, Kalle}}, booktitle = {{2024 IEEE 18th International Conference on Automatic Face and Gesture Recognition, FG 2024}}, isbn = {{9798350394948}}, language = {{eng}}, month = {{07}}, publisher = {{IEEE - Institute of Electrical and Electronics Engineers Inc.}}, title = {{Geometry-Biased Transformer for Robust Multi-View 3D Human Pose Reconstruction}}, url = {{http://dx.doi.org/10.1109/FG59268.2024.10581930}}, doi = {{10.1109/FG59268.2024.10581930}}, year = {{2024}}, }