Triplet Prediction on 3DSSG
66.58mA@50SGFN
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SGFNBackbone=PointTransformer, Initialization=ToLL, Fine-tuning Protocol=Full fine-tuning2026.03 | 66.58 | 74.32 | |
| VL-SATBackbone=PointTransformer, Initialization=ToLL, Fine-tuning Protocol=Full fine-tuning2026.03 | 66.42 | 75.59 | |
| VL-SATBackbone=PointNet, Initialization=Learning from scratch, Fine-tuning Protocol=Full fine-tuning2026.03 | 65.09 | 73.59 | |
| VL-SATBackbone=PointTransformer, Initialization=PointDif, Fine-tuning Protocol=Full fine-tuning2026.03 | 64.86 | 73.66 | |
| OCRLBackbone=PointTransformer, Initialization=Learning from scratch, Fine-tuning Protocol=Full fine-tuning2026.03 | 63.79 | 73.48 | |
| SGFNBackbone=PointTransformer, Initialization=PointDif, Fine-tuning Protocol=Full fine-tuning2026.03 | 63.61 | 72.89 | |
| VL-SATBackbone=PointTransformer, Initialization=Learning from scratch, Fine-tuning Protocol=Full fine-tuning2026.03 | 63.55 | 72.48 | |
| OCRLBackbone=PointNet, Initialization=Learning from scratch, Fine-tuning Protocol=Full fine-tuning2026.03 | 62.29 | 70.26 | |
| SGFNBackbone=PointTransformer, Initialization=Learning from scratch, Fine-tuning Protocol=Full fine-tuning2026.03 | 60.05 | 70.26 | |
| SGFNBackbone=PointNet, Initialization=Learning from scratch, Fine-tuning Protocol=Full fine-tuning2026.03 | 58.37 | 67.61 | |
| SGPNBackbone=PointNet, Initialization=Learning from scratch, Fine-tuning Protocol=Full fine-tuning2026.03 | 52.74 | 65.58 | |
| VL-SATBackbone=PointTransformer, Initialization=ToLL, Fine-tuning Protocol=MLP-only2026.03 | 50.05 | 60.65 | |
| SGFNBackbone=PointTransformer, Initialization=ToLL, Fine-tuning Protocol=MLP-only2026.03 | 48.87 | 58.29 | |
| VL-SATBackbone=PointTransformer, Initialization=Random Parameters, Fine-tuning Protocol=MLP-only2026.03 | 40.13 | 51.03 | |
| SGFNBackbone=PointTransformer, Initialization=Random Parameters, Fine-tuning Protocol=MLP-only2026.03 | 39.97 | 49.11 |