Downloads · 30 days
0
Cloth-splatters/dexgarmentlab-lift-20260822-state-est-gps-flow-tf2
dexgarmentlab-lift-20260822-state-est-gps-flow-tf2 is a machine learning model from Cloth-splatters. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for diffusers. The card lists the license as mit.
Superseded (2026-09-16) by dexgarmentlab-lift-20260822-state-est-gps-flow-tf2-noscale for the paper's numbers. Revision 03e61682 (tag paper-figures-ep20) rendered the Fig. 1 and Fig. 2 panels of the submission and sta…
Downloads · 30 days
0
Access
Public
Updated Sep 16, 2026
Repo size
91 MB
Likes
0
Public
Click a slice to open those files.
.safetensors18.2 MB · 100%
From the Hugging Face model README
Superseded (2026-09-16) by dexgarmentlab-lift-20260822-state-est-gps-flow-tf2-noscale for the paper's numbers. Revision
03e61682(tagpaper-figures-ep20) rendered the Fig. 1 and Fig. 2 panels of the submission and stays available. Current release: the ClothAtlas collection.
Flow-matching GPSStateEstModel trained on the full-state 2026-08-22 lift
collection. It reconstructs the variable-topology cloth mesh from visible
point clouds and uses two-frame temporal conditioning (num_context_frames: 2,
single-frame at inference via the pcd_prev=None fallback). model/ contains
the best EMA checkpoint.
dexgarmentlab_lift_full_state_20260822.h5GPSStateEstModel, temporal context 2, coord_scale 3mesh-hypothesis-pfLoad through UniClothDiff's src.hub.resolve_checkpoint
(dexgarmentlab-lift-20260822-state-est-gps-flow-tf2). The complete training
recipe is config.yml; the Diffusers-compatible model is under model/.
The lift best recipe (v2 EMA + heavy augmentation + two-frame temporal +
template-size mismatch + coord_scale 3) plus orientation augmentation: joint random yaw
about gravity (random_z_rotation: true) and template-alone yaw mismatch
(template_yaw_max_deg: 180), so template yaw stops being ground truth and
in-plane orientation must be read from the observation. Measured on the real
take-2 pants with no alignment at all: flat orientation error 41–55° → 2–11°,
symmetric Chamfer 12.55 → 8.77 mm (best take-2 arm of the campaign); the
PCA pre-alignment step is unnecessary for this revision. Cost: unaugmented
validation loss 1.10e-4 m² vs 8.14e-5 (arm 17, job 960947_17, tag noyaw) for the non-yaw recipe, and near-symmetric
flat silhouettes can resolve to the wrong 180° facing (the template-facing
prior is deleted; use the orientation tie-break, or wait for the mixture-prior
template_yaw_p 0.5 arms).
dexgarment_flow_v2_cs3_tf2_lift20260822_yaw_2026-08-31_21-58-39_976630 (checkpoint-best, EMA)976630_19 (COMPLETED, 0:0)0.0001104726565322792578860d317dab28fb689a956d30b5d64ee4e6d7122f22f99bc3af4de3765b5147The yaw recipe of the 2026-09-01 release with a mixture prior on the template
yaw (template_yaw_p: 0.5: half the training samples keep the exact template
yaw, the other half get a uniform random one), i.e. the complete training
randomization of the paper: joint random scale + yaw of cloud, target and rest
mesh; template-alone scale mismatch (log-uniform 0.5-2) and yaw mismatch with
probability 0.5; point jitter, dropout, wedge dropout and outliers. This is the
checkpoint behind every simulation result in the paper.
Measured on the garment-disjoint lift test split (22 garments, 221-mesh library): flat-frame canonical-mesh retrieval 21/22 rank 1 and 22/22 top-5, median rank 2 over 198 observations; energy-selected vertex error 5.7 mm on flat garments and 17.4 mm over all 7,537 test frames; error flat within 1 mm across template scale factors 0.5-2 and all template yaw offsets. Reconstruction saturates at 3-5 denoising steps (15.8 mm at 5 steps vs 17.2 at 30) — use 5. Real take-2 pants, no alignment: 8.35 mm symmetric Chamfer, a tie with the 2026-09-01 yaw release; the 180° facing ambiguity of near-symmetric flat silhouettes remains.
dexgarment_flow_v2_cs3_tf2_lift20260822_yawmix_2026-09-01_08-35-23_982082 (checkpoint-best, EMA)982082_21 (COMPLETED, 0:0)0.00010988347423944386711dfa6034c954d98a449ad7b1d414f9b323814352f4b642662c6634a9094209d19d718d001b9c3550d5d7d695586f72079b2134The yaw-mixture flow recipe of the 2026-09-03 release trained for 20 epochs
instead of 10 (fresh run, cosine horizon doubled; job 1134838_21). Same data,
randomization and seed. This is the checkpoint behind every flow-matching
result in the paper.
Measured on the garment-disjoint lift test split (22 garments, 221-mesh library), against the 10-epoch release: energy-selected vertex error 15.8 mm over all 7,537 test frames (was 17.3), 5.8 mm on flat garments (was 5.5); symmetric surface CD 11.7 mm (was 12.2), 12.0 mm on the camera-occluded surface (was 12.5); few-step saturation 15.0 mm at 5 steps and 16.3 at 30 (was 15.8 / 17.3), one-step 27.5 mm (was 36.3). Canonical-mesh retrieval: flat frame 22/22 rank 1 (unchanged), one crumpled observation at frame 200 12/22 rank 1 and 18/22 top-5 (was 10/22 and 14/22), 120/198 rank-1 single-observation decisions (was 107/198). Validation loss min 8.5e-5 (was 1.1e-4) and stable at 0.9-1.1e-4 over the last quarter of training.
dexgarment_flow_v2_cs3_tf2_lift20260822_yawmix_2026-09-08_14-37-03_1134838 (checkpoint-best, EMA)1134838_21 (COMPLETED, 0:0)8.472606263898478e-0533a60c6c989dac22f4934f264c8d8da6936e5ef2c542565be511ba0d2c13cd881a5d02855cd2bd4e66008720f637a276d887796e