SapiensID: Foundation for Human Recognition
Paper • 2504.04708 • Published
SapiensID (wb4m) — JAX/Equinox weights
Converted weights for SapiensIdEqx, a JAX/Equinox re-implementation of SapiensID: Foundation for Human Recognition (Kim, Ye, Su, Liu and Liu, CVPR 2025, arXiv:2504.04708, https://github.com/mk-minchul/sapiensid).
Licence — NON-COMMERCIAL
These weights are a format conversion of the authors' released wb4m checkpoint and remain under
Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) —
https://creativecommons.org/licenses/by-nc/4.0/ (full text in LICENSE, copied verbatim from the
upstream repository). NonCommercial restricts USE, not only redistribution. Research and other
non-commercial use only.
Attribution is owed to the SapiensID authors, and — for the dynamic-patching masking core, which derives from Meta's MAE — to:
Copyright (c) Meta Platforms, Inc. and affiliates. All rights reserved.
THIRD_PARTY_NOTICES.md records every component's provenance and licence.
Unless otherwise separately undertaken by the Licensor, to the extent possible, the Licensor offers the Licensed Material as-is and as-available, and makes no representations or warranties of any kind concerning the Licensed Material, whether express, implied, statutory, or other.
What this file is
| file | sapiensid_wb4m.eqx (Equinox tree_serialise_leaves) |
| sha256 | 620c405bb50bd82c868fe3ae9bf35731d8ccf7a526227ab3b299b3988ae0adc6 |
| converted from | upstream wb4m model.pth, 1,646,175,550 bytes, sha256 8df1e2d7a64fbc87969326862aeafccc83eb5ee0462a4f007d7f09603782cf02, crc32 941cf66d |
| equivalence | all 297 array leaves bit-identical to the model built by SapiensId.from_pytorch from that checkpoint |
The port reproduces the reference forward to 1 − cos 1.79e-12 on the 4096-d embedding, gated at 56 seams end to end on CPU and GPU. Inputs are an image plus caller-supplied body keypoints and foreground mask; the upstream aligner (YOLOv8 / OpenPose) is not part of this port.
Use
from sapiensid_eqx.model import SapiensId
model = SapiensId.from_pretrained("sapiensid_wb4m.eqx")
No PyTorch is needed to load or run these weights.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support