What does MPJPE measure, and which public monocular pose benchmarks actually use marker-based optical ground truth?
Corrected 2026-10-24 - see C3-B1090-01; the original statements below are kept for history and are superseded where the correction says so. MPJPE is the mean Euclidean distance (usually in mm) between predicted and reference 3D joint positions, with P-MPJPE reporting the error after a rigid-plus-scale alignment. Of the common benchmarks, only Human3.6M (10-camera Vicon MX T40 at 200 Hz, with reflective markers) is marker-based optical; 3DPW uses body-worn inertial sensors and EMDB uses electromagnetic sensors, and EMDB's authors estimate their own references at about 2.3-2.6 cm PA-MPJPE. So "MPJPE against marker-based ground truth" is a Human3.6M-style statement and is not interchangeable with in-the-wild numbers.
MPJPE is the mean Euclidean distance (usually in mm) between predicted and reference 3D joint positions, with P-MPJPE reporting the error after a rigid-plus-scale alignment. Of the common benchmarks, only Human3.6M (10-camera Vicon MX T40 at 200 Hz, with reflective markers) is marker-based optical; 3DPW uses body-worn inertial sensors and EMDB uses electromagnetic sensors, and EMDB's authors estimate their own references at about 2.3-2.6 cm PA-MPJPE. So "MPJPE against marker-based ground truth" is a Human3.6M-style statement and is not interchangeable with in-the-wild numbers.
Curator correction 2026-10-24 (discovery audit record C3-B1090-01, curator-reviewed C3R-0696; excerpts re-located by independent curator GET 2026-10-24): Correction. Record C3-B044-W10-06 summary says 'EMDB's authors estimate their own references at about 2.3-2.6 cm PA-MPJPE'. In the EMDB paper (arXiv 2308.16894v1, submitted 31 Aug 2023) only 2.3 cm (10.6 degrees angular) is EMDB's own estimate; its Table 2 gives 2.6 cm (12.1 degrees) as the accuracy figure for the other dataset, 3DPW. The record's findings row names the two values correctly; only the summary sentence merges them. The summary was inaccurate as written, not made stale by a later version (the v1 PDF was re-fetched 2026-10-24). Separately, the record's evidence row E5 cites the Human3.6M description page for the 'MX System: Vicon T40, 10 cameras, 200Hz' table, but in the fetched HTML that table sits inside an HTML comment and is not displayed; the same hardware figures are printed in the Human3.6M PAMI paper's dataset table, which is the better citation.
Curator qualifications: All excerpts located; MPJPE definition and which benchmarks use optical ground truth.
Evidence status: statements located in the cited sources by a correlated desk review (Claude); not executed; not independently reviewed.
Structured fields
{
"citations": [
{
"url": "https://arxiv.org/pdf/1811.11742v2",
"note": "candidate evidence excerpts"
},
{
"url": "http://vision.imar.ro/human3.6m/pami-h36m.pdf",
"note": "candidate evidence excerpts"
},
{
"url": "http://vision.imar.ro/human3.6m/description.php",
"note": "candidate evidence excerpts"
},
{
"url": "https://arxiv.org/pdf/2312.07531v2",
"note": "candidate evidence excerpts"
},
{
"url": "https://arxiv.org/pdf/2308.16894v1",
"note": "candidate evidence excerpts"
}
]
}Evidence claims
- [sourced_fact] Curator correction 2026-10-24 (discovery audit record C3-B1090-01, curator-reviewed C3R-0696; excerpts re-located by independent curator GET 2026-10-24): Correction. Record C3-B044-W10-06 summary says 'EMDB's authors estimate their own references at about 2.3-2.6 cm PA-MPJPE'. In the EMDB paper (arXiv 2308.16894v1, submitted 31 Aug 2023) only 2.3 cm (10.6 degrees angular) is EMDB's own estimate; its Table 2 gives 2.6 cm (12.1 degrees) as the accuracy figure for the other dataset, 3DPW. The record's findings row names the two values correctly; only the summary sentence merges them. The summary was inaccurate as written, not made stale by a later version (the v1 PDF was re-fetched 2026-10-24). Separately, the record's evidence row E5 cites the Human3.6M description page for the 'MX System: Vicon T40, 10 cameras, 200Hz' table, but in the fetched HTML that table sits inside an HTML comment and is not displayed; the same hardware figures are printed in the Human3.6M PAMI paper's dataset table, which is the better citation. · locator: curator correction paragraph in this record body (C3-B1090-01) source
- [sourced_fact] SUPERSEDED IN PART 2026-10-24 by curator correction C3-B1090-01 (see the correction claim above); original kept for history: MPJPE is the mean Euclidean distance (usually in mm) between predicted and reference 3D joint positions, with P-MPJPE reporting the error after a rigid-plus-scale alignment. Of the common benchmarks, only Human3.6M (10-camera Vicon MX T40 at 200 Hz, with reflective markers) is marker-based optical; 3DPW uses body-worn inertial sensors and EMDB uses electromagnetic sensors, and EMDB's authors estimate their own references at about 2.3-2.6 cm PA-MPJPE. So "MPJPE against marker-based ground truth" is a Human3.6M-style statement and is not interchangeable with in-the-wild numbers. · locator: candidate evidence excerpts source
Rights
| component | license | hosting | redistribution | modification | training | Evidence and notes |
|---|---|---|---|---|---|---|
| documentation | unknown | permitted | permitted | unknown | unknown | Applies to this original summary only, not to the cited sources. |