Models trained on synthetic classroom incident data transfer zero-shot to real-world footage, with the proposed method best at 63.41% though a synthetic-to-real domain gap remains¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? In zero-shot synthetic-to-real transfer, the proposed method obtains the best real-world accuracy of 63.41%, outperforming MSG3D by 4.18 percentage points and other baselines by larger margins. → Paritosh Parmar 2026 q2 i? All methods drop in accuracy from synthetic to real-world data, so synthetic data does not fully replace real-world data. → Paritosh Parmar 2026 (2)

Evidence¶

Paritosh Parmar 2026¶

Paritosh Parmar, Landy Lan, Hong Yang, Chen Yi, Chiat Pin Tay. (2026). Robust and Efficient Motion Reasoning for Privacy-Aware Classroom Incident Recognition. arXiv preprint. https://arxiv.org/abs/2608.05115 doi:10.48550/arxiv.2608.05115

q2 · i? · design · r2

Zero-shot transfer experiment: models trained only on the synthetic set, evaluated directly on the 574-sample real-world test set without fine-tuning. The proposed method "obtains the best real-world accuracy of 63.41%"; accuracies only, no effect size.

"However, our method obtains the best real-world accuracy of 63.41%, outperforming MSG3D, the strongest baseline in this setting, by 4.18 percentage points."

Paritosh Parmar 2026 (2)¶

Paritosh Parmar, Landy Lan, Hong Yang, Chen Yi, Chiat Pin Tay. (2026). Robust and Efficient Motion Reasoning for Privacy-Aware Classroom Incident Recognition. arXiv preprint. https://arxiv.org/abs/2608.05115

q2 · i? · design · r2

Same zero-shot evaluation, reported as the authors' caveat on interpreting the transfer results: the drop from 71.78% to 63.41% "shows that a synthetic-to-real domain gap remains", so synthetic data provides a useful training signal but does not replace real data.

"the decrease from 71.78% on synthetic data to 63.41% on real-world data shows that a synthetic-to-real domain gap remains."

Discussion¶

Related Claims¶

• Classroom incident recognition is substantially harder than established general action-recognition benchmarks, with PoseC3D dropping from over 94% accuracy to 70.54% — related
• A distilled multi-order kinematic model outperforms pose-based baselines on a synthetic classroom incident dataset while using one-tenth as many parameters as PoseC3D — related
• The framework is robust to 15% ASR transcription noise and retains 99.6% of its ICS under zero-shot cross-domain transfer to EdNet-Math — related
• Distilling a multi-order hierarchical teacher into a smaller zeroth-order student lets the student outperform the teacher, and improves 4.1 percentage points in accuracy over STGCN++