cs.CVAug 20, 2026
SaveAggregate, Don't Adapt: Subject-Level Posterior Aggregation and Transductive Calibration for Cross-Site Parkinsonian Gait Severity
Organizations: University of Alberta, Edmonton, Canada
Abstract
We describe the winning entry to the MoCha 2026 Benchmark and Challenge on Parkinsonian Gait, which predicts MDS-UPDRS gait severity from canonicalized SMPL motion recorded at clinical sites unseen during training. The system reaches 0.6945 macro-F1 on the hidden test and ranked first of 58 entries, ahead of the runner-up at 0.5807 and the organizers' baseline at 0.4289, on a frozen public motion encoder with a single linear layer. Nearly all of the margin comes from three stages usually treated as bookkeeping: reproducing the reference benchmark's exact head recipe, averaging per-walk posteriors within the subject grouping the organizers ship, and a label-free transductive calibration of the feature mean and the decision operating point. Fine-tuning the encoder lost in four distinct forms, and ten alternative encoders were worse. Every ablation number is a paid read on the hidden test, because our own leave-two-cohort-out cross-validation proved anti-correlated with the deciding score over eleven configurations. We give the negative record in full, and identify our largest gain, subject-level aggregation, as the binding ceiling on this benchmark.
Figures & tables
Figure 1: The winning margin came from reading the input format, not from the motion representation. The encoder is public and frozen; the only trained parameters are a 14 KB linear layer (top). Averaging per-walk posteriors within the subject grouping that the organizers ship inside the input is worth macro-F1, more than every representation change we made combined (bottom); every bar is a paid read on the hidden test. That same aggregation is also this benchmark’s ceiling ( Sec. 7 ).
Figure 2: Severity is labeled per walk, and a subject’s walks do not have to agree. Four walks of one T-SDU-PD subject, drawn from the released canonicalized SMPL motion through forward kinematics and projected to the side view the encoder reads; three poses are shown per walk. Two carry ground-truth class 1 and two carry class 2. The strip gives all eleven walks of that subject. Averaging over this grouping is what buys macro-F1 ( Sec. 5 ); the residual disagreement inside it is what caps the system ( Sec. 7 ).
| cohort | subjects | walks | 0 | 1 | 2 | 3 | mixed |
|---|---|---|---|---|---|---|---|
| 3DGait | 43 | 90 | 24 | 42 | 14 | 10 | 16 |
| PD-GaM | 30 | 1700 | 783 | 635 | 248 | 34 | 29 |
| BMCLab | 23 | 781 | 341 | 276 | 164 | 0 | 8 |
| T-SDU-PD | 14 | 381 | 96 | 118 | 167 | 0 | 9 |
| all four | 110 | 2952 | 1244 | 1071 | 593 | 44 | 62 |
Table 1: The four UPDRS-labeled CARE-PD cohorts, counted from the released pickles. Columns 0–3 count walks per severity class, and mixed counts subjects whose own walks do not all carry the same label. Class 3 is rare and missing from two cohorts, and more than half of all subjects are mixed.
Figure 3: Our local cross-validation ranked configurations backwards. Each point is one of the eleven configurations for which we hold both a leave-two-cohort-out score and a paid hidden-test read. Filled points use full subject aggregation, open points use less. The inversion is concentrated on the aggregation axis, which is also the axis worth the most.
| # | team | macro-F1 | macro-P | macro-R | accuracy | QWK |
|---|---|---|---|---|---|---|
| 1 | JLShen (ours) | 0.69447 | 0.72347 | 0.67659 | 0.65591 | 0.60007 |
| 2 | brady_kinesia | 0.58070 | 0.64650 | 0.55449 | 0.53495 | 0.40674 |
| 3 | unist_visionlab | 0.57358 | 0.56464 | 0.58659 | 0.54301 | 0.42932 |
| 4 | Nottingham_RVCE | 0.55577 | 0.59830 | 0.53448 | 0.53763 | 0.42050 |
| 5 | tuananh1007 | 0.54907 | 0.59898 | 0.52355 | 0.48925 | 0.43321 |
| 6 | anhnamxtanh | 0.54097 | 0.58588 | 0.51573 | 0.50806 | 0.40438 |
Table 2: Final standing on the MoCha 2026 hidden test. Every column is the organizers’ scorer on the hidden test; our ranked entry is Codabench submission 882979. Leaderboard 18564, phase 27428, read at the close of the phase on 2026-08-16, when the board carried 58 entries. Boards stay editable after a phase closes: a re-read on 2026-08-20 listed 50 entries with this ordering unchanged.
| stage | macro-F1 | |
|---|---|---|
| organizers’ released baseline bundle | 0.4289 | — |
| frozen MotionAGFormer-S + plain linear probe | 0.467 | |
| + benchmark-exact head recipe (focal loss, AdamW, z-score) | 0.524 | |
| + transductive centering & -divisor operating point | 0.5407 | |
| + subject-mean posterior aggregation ( ) | 0.68371 | |
| + centering shrinkage retuned | 0.68699 |
Table 3: Component ladder. Every row is a paid read of the organizers’ scorer on the hidden test, never a local estimate. Rows were measured at different points in the run, so each is against the row above it on the pipeline as it then stood, not a strictly nested leave-one-out ablation.
| family | outcome |
|---|---|
| Encoder fine-tuning | Dead in four forms: unsupervised 2D 3D pretext fine-tuning (server 0.3605, far below frozen), anti-site-latch low-rank adaptation (LoRA) with gradient reversal and a bottleneck, supervised LoRA, and contrastive self-supervision with a gradient reversal layer. |
| Alternative encoders | All below MotionAGFormer-S: MixSTE [ 25 ] (server 0.4081, including in its own native normalization pipeline), MotionAGFormer-B/L, MotionBERT [ 27 ] , PoseFormerV2, MotionCLIP, TMR, an NTU-pretrained skeleton transformer, MoMask, a probe on the Riemannian manifold of symmetric positive-definite matrices, and a time-series foundation model. |
| Alignment beyond the first moment | Only the mean transfers. CORAL [ 22 ] and diagonal-std alignment, median centering, PCA denoising, iterative nullspace projection, transductive batch normalization, quantile-CDF matching, and per-site clustered centering all lost. The last lost even with oracle site labels , so it is not a clustering-quality problem. |
| Decision rules and objectives | Ordinal and CORAL-style ordinal heads (server 0.4558, , bracketed over so not a mistuning artifact), threshold tuning, linear discriminant analysis, Saerens EM label shift [ 17 ] , class-balanced centering, prototype and nearest-class-mean rules, a soft macro-F1 surrogate objective, the invariance objectives V-REx, GroupDRO [ 18 ] and IRM, and training-time de-confidencing. A parameter-free joint macro-F1 decoder returned predictions byte-identical to argmax on the server, so argmax is already F1-optimal here , and the recall gap is separability rather than decision-rule miscalibration. Per-class decision scaling also lost in both directions: suppressing class 3 gave 0.64882 and boosting class 2 gave 0.5715. |
| Ensembles | Server-negative in every form: diverse-encoder blends, MotionAGFormer MixSTE (locally positive, server collapse to 0.3720), fine-tuned frozen, and two-view side and back (cross-validation inverted to on the server). See the note below. |
| Transductive inference beyond centering | Graph label propagation [ 26 ] , Nyström-RBF and metric learning, density-ratio importance weighting, FixMatch [ 20 ] over the five unlabeled cohorts, Deep Feature Reweighting [ 9 ] , attention pooling over clips, bag-of-windows pooling, and self-training test-time adaptation [ 23 ] . The last was an exact no-op, changing 0 of 284 probe items. |
Table 4: Rejected families, each closed by measurement. “Server” marks a paid read of the organizers’ scorer on the hidden test; the remaining entries are local cross-validation, which Sec. 4 shows is a weak selector on decision and aggregation axes but a usable one for gross representation failure.
Figure 4: One label per subject is worth , and it is also a hard bound. Left: most subjects in the labeled cohorts carry more than one severity label across their own walks. Right: predicting one label per subject can reach at most 69.5% walk accuracy on those cohorts, and the deployed system realizes 65.6% on the hidden test. Both panels count released labels only.
More in this topic
Gait Analysis
Gait analysis uses movement and biomechanical signals to characterize walking, recognize people, assess health, and guide rehabilitation or locomotion.
See all 75 papers