Zero-shot Segmentation on MS-COCO
49.9mAPBase
Evaluation Results
| Method | Links | |
|---|---|---|
| BaseSAM Backbone=SAM-H, Density=100%, Speedup=× 1.00, Detector=DINO2026.05 | 49.9 | |
| SparseSAMSAM Backbone=SAM-H, Density=50%, Speedup=× 1.22, Compression Type=attention-only, Detector=DINO2026.05 | 49.9 | |
| PieceWise AttentionSAM Backbone=SAM-H, Density=50%, Speedup=× 0.65, Compression Type=attention-only, Detector=DINO2026.05 | 49.8 | |
| ToMeSAM Backbone=SAM-H, Density=50%, Speedup=× 0.64, Compression Type=attention+MLP, Detector=DINO2026.05 | 49.8 | |
| BaseSAM Backbone=SAM-L, Density=100%, Speedup=× 1.00, Detector=DINO2026.05 | 49.5 | |
| PieceWise AttentionSAM Backbone=SAM-L, Density=50%, Speedup=× 0.67, Compression Type=attention-only, Detector=DINO2026.05 | 49.4 | |
| ToMeSAM Backbone=SAM-L, Density=50%, Speedup=× 0.75, Compression Type=attention+MLP, Detector=DINO2026.05 | 49.4 | |
| SparseSAMSAM Backbone=SAM-L, Density=50%, Speedup=× 1.57, Compression Type=attention-only, Detector=DINO2026.05 | 49.4 | |
| SparseSAMSAM Backbone=SAM-H, Density=25%, Speedup=× 1.28, Compression Type=attention-only, Detector=DINO2026.05 | 49.4 | |
| Sparge AttentionSAM Backbone=SAM-L, Density=50%, Speedup=× 1.21, Compression Type=attention-only, Detector=DINO2026.05 | 49.1 | |
| SparseSAMSAM Backbone=SAM-L, Density=25%, Speedup=× 1.64, Compression Type=attention-only, Detector=DINO2026.05 | 48.7 | |
| SparseSAMSAM Backbone=SAM-H, Density=50%, Speedup=× 1.40, Compression Type=attention+MLP, Detector=DINO2026.05 | 48.7 | |
| PieceWise AttentionSAM Backbone=SAM-H, Density=25%, Speedup=× 0.69, Compression Type=attention-only, Detector=DINO2026.05 | 48.6 | |
| PieceWise AttentionSAM Backbone=SAM-L, Density=25%, Speedup=× 0.68, Compression Type=attention-only, Detector=DINO2026.05 | 48.3 | |
| SparseSAMSAM Backbone=SAM-L, Density=50%, Speedup=× 1.83, Compression Type=attention+MLP, Detector=DINO2026.05 | 48.2 | |
| SparseSAMSAM Backbone=SAM-H, Density=25%, Speedup=× 1.59, Compression Type=attention+MLP, Detector=DINO2026.05 | 47.2 | |
| BaseSAM Backbone=SAM-B, Density=100%, Speedup=× 1.00, Detector=DINO2026.05 | 46.8 | |
| SparseSAMSAM Backbone=SAM-L, Density=25%, Speedup=× 2.05, Compression Type=attention+MLP, Detector=DINO2026.05 | 46.8 | |
| ToMeSAM Backbone=SAM-B, Density=50%, Speedup=× 0.74, Compression Type=attention+MLP, Detector=DINO2026.05 | 46.7 | |
| SparseSAMSAM Backbone=SAM-B, Density=50%, Speedup=× 1.55, Compression Type=attention-only, Detector=DINO2026.05 | 46.7 | |
| PieceWise AttentionSAM Backbone=SAM-B, Density=50%, Speedup=× 0.72, Compression Type=attention-only, Detector=DINO2026.05 | 46.5 | |
| Sparge AttentionSAM Backbone=SAM-B, Density=50%, Speedup=× 1.24, Compression Type=attention-only, Detector=DINO2026.05 | 46.3 | |
| Sparge AttentionSAM Backbone=SAM-L, Density=25%, Speedup=× 1.23, Compression Type=attention-only, Detector=DINO2026.05 | 46.1 | |
| SparseSAMSAM Backbone=SAM-B, Density=25%, Speedup=× 1.63, Compression Type=attention-only, Detector=DINO2026.05 | 45.9 | |
| SparseSAMSAM Backbone=SAM-B, Density=50%, Speedup=× 1.89, Compression Type=attention+MLP, Detector=DINO2026.05 | 45.9 | |
| PieceWise AttentionSAM Backbone=SAM-B, Density=25%, Speedup=× 0.73, Compression Type=attention-only, Detector=DINO2026.05 | 45.5 | |
| StructSAMSAM Backbone=SAM-H, Density=50%, Speedup=× 0.79, Compression Type=attention+MLP, Detector=DINO2026.05 | 45.1 | |
| SparseSAMSAM Backbone=SAM-B, Density=25%, Speedup=× 2.15, Compression Type=attention+MLP, Detector=DINO2026.05 | 45.1 | |
| Sparge AttentionSAM Backbone=SAM-B, Density=25%, Speedup=× 1.25, Compression Type=attention-only, Detector=DINO2026.05 | 44.7 | |
| StructSAMSAM Backbone=SAM-L, Density=50%, Speedup=× 0.74, Compression Type=attention+MLP, Detector=DINO2026.05 | 44.5 | |
| ToMeSAM Backbone=SAM-H, Density=25%, Speedup=× 0.80, Compression Type=attention+MLP, Detector=DINO2026.05 | 43 | |
| ToMeSAM Backbone=SAM-L, Density=25%, Speedup=× 0.86, Compression Type=attention+MLP, Detector=DINO2026.05 | 42.5 | |
| StructSAMSAM Backbone=SAM-B, Density=50%, Speedup=× 0.71, Compression Type=attention+MLP, Detector=DINO2026.05 | 42.2 | |
| ToMeSAM Backbone=SAM-B, Density=25%, Speedup=× 0.83, Compression Type=attention+MLP, Detector=DINO2026.05 | 40.3 | |
| StructSAMSAM Backbone=SAM-L, Density=25%, Speedup=× 0.89, Compression Type=attention+MLP, Detector=DINO2026.05 | 31.8 | |
| StructSAMSAM Backbone=SAM-H, Density=25%, Speedup=× 0.94, Compression Type=attention+MLP, Detector=DINO2026.05 | 31.8 | |
| StructSAMSAM Backbone=SAM-B, Density=25%, Speedup=× 0.86, Compression Type=attention+MLP, Detector=DINO2026.05 | 30.6 |