Gesture Recognition on MultiMeDaLIS (5-fold cross-validation train)
80.5Top-1 AccuracyCAST full (asymmetric fusion)
Evaluation Results
| Method | Links | |
|---|---|---|
| CAST full (asymmetric fusion)Components=CAST, Fusion=Asymmetric, Protocol=5-fold cross-validation2026.05 | 80.5 | |
| + CVD + CASA (concat fusion)Components=CVD + CASA, Fusion=Concat, Protocol=5-fold cross-validation2026.05 | 79.3 | |
| 6-channel RTM+pseudo-RDM (Swin-S)Backbone=Swin-S, Protocol=5-fold cross-validation2026.05 | 78.6 | |
| + CVD only (no CASA, concat fusion)Components=CVD, Fusion=Concat, Protocol=5-fold cross-validation2026.05 | 78.1 | |
| + CASA only (no CVD)Components=CASA, Protocol=5-fold cross-validation2026.05 | 77.9 | |
| 6-channel RTM+pseudo-RDM (ViT-S)Backbone=ViT-S, Protocol=5-fold cross-validation2026.05 | 77.4 | |
| EfficientNetV2-S (single fold)Backbone=EfficientNetV2-S, Protocol=5-fold cross-validation2026.05 | 77.2 | |
| ConvNeXt-Tiny (single fold)Backbone=ConvNeXt-Tiny, Protocol=5-fold cross-validation2026.05 | 76.8 |