Papers › A generic diffusion-based approach for 3D human pose prediction in the wild

A generic diffusion-based approach for 3D human pose prediction in the wild

11 Oct 2022arXiv:2210.05669archive 2025-07-28

Saeed Saadatnejad, Ali Rasekh, Mohammadreza Mofayezi, Yasamin Medghalchi, Sara Rajabzadeh, Taylor Mordan, Alexandre Alahi

Predicting 3D human poses in real-world scenarios, also known as human pose forecasting, is inevitably subject to noisy inputs arising from inaccurate 3D pose estimations and occlusions. To address these challenges, we propose a diffusion-based approach that can predict given noisy observations. We frame the prediction task as a denoising problem, where both observation and prediction are considered as a single sequence containing missing elements (whether in the observation or prediction horizon). All missing elements are treated as noise and denoised with our conditional diffusion model. To better handle long-term forecasting horizon, we present a temporal cascaded diffusion model. We demonstrate the benefits of our approach on four publicly available datasets (Human3.6M, HumanEva-I, AMASS, and 3DPW), outperforming the state-of-the-art. Additionally, we show that our framework is generic enough to improve any 3D pose prediction model as a pre-processing step to repair their inputs and a post-processing step to refine their outputs. The code is available online: \url{https://github.com/vita-epfl/DePOSit}.

PaperPDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

vita-epfl/deposit officialmentioned in papermentioned on GitHubpytorchAGPL-3.0 report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

DenoisingHuman Pose ForecastingMissing ElementsPose PredictionPrediction

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Human Pose Forecasting 3DPW TCD FDE@1000ms (mm) 73.4 #7 of 7 Archive leaderboard report
Human Pose Forecasting 3DPW TCD FDE@560ms (mm) 55.4 #7 of 7 Archive leaderboard report
Human Pose Forecasting 3DPW TCD FDE@720ms (mm) 61.6 #7 of 7 Archive leaderboard report
Human Pose Forecasting 3DPW TCD FDE@880ms (mm) 67.9 #7 of 7 Archive leaderboard report
Human Pose Forecasting AMASS TCD FDE@1000ms (mm) 66.7 #6 of 11 Archive leaderboard report
Human Pose Forecasting AMASS TCD FDE@560ms (mm) 49.8 #6 of 11 Archive leaderboard report
Human Pose Forecasting AMASS TCD FDE@720ms (mm) 54.5 #6 of 11 Archive leaderboard report
Human Pose Forecasting AMASS TCD FDE@880ms (mm) 60.1 #6 of 11 Archive leaderboard report
Human Pose Forecasting Human3.6M TCD ADE 356 #22 of 33 Archive leaderboard report
Human Pose Forecasting Human3.6M TCD APD 19466 #22 of 33 Archive leaderboard report
Human Pose Forecasting Human3.6M TCD FDE 396 #22 of 33 Archive leaderboard report
Human Pose Forecasting Human3.6M TCD MMADE 463 #22 of 33 Archive leaderboard report
Human Pose Forecasting Human3.6M TCD MMFDE 445 #22 of 33 Archive leaderboard report
Human Pose Forecasting HumanEva-I TCD ADE@2000ms 199 #1 of 11 Archive leaderboard report
Human Pose Forecasting HumanEva-I TCD APD@2000ms 6764 #1 of 11 Archive leaderboard report
Human Pose Forecasting HumanEva-I TCD FDE@2000ms 215 #1 of 11 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

DiffusionRepair

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections