A low-rank probe recovered motion-state information from frozen 3D features
Source check: Source checked. Not independent scientific confirmation.
Finding: A fitted rank-eight probe on frozen π³X features reached 91.67% balanced accuracy when classifying whether token pairs shared a motion state. Shuffling training labels reduced balanced accuracy to 50.34%. The result is specific to this probe and these recordings.
Technology: π³X; fitted low-rank motion probe
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
- AI contribution
- Matched and shuffled motion labels trained low-rank quadratic probes on frozen π³X visual tokens.
- Test and result
- The test compared probes with matched or within-frame shuffled targets on YouTube-VOS validation pairs.
- Limitation / still unknown
- Semantic content and object identity can contribute to the signal. The test does not establish an intrinsic motion basis.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://arxiv.org/abs/2610.33802v1 · https://arxiv.org/html/2610.33802v1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 21 Nov 2026. Review scope: v1 rank-restricted probing, Table 5 and Appendix D.3.
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.