Class-Incremental Learning on ObjNet B0 Inc10
72.16Avg AccuracyRanpac + C-Flat Turbo
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ranpac + C-Flat TurboModel Category=PTM-based, Backbone=ViT-B/16-IN1K, Throughput=94.34 (61.0%)2026.04 | 72.16 | 60.33 | |
| Ranpac + C-FlatModel Category=PTM-based, Backbone=ViT-B/16-IN1K, Throughput=42.98 (27.8%)2026.04 | 72.15 | 60.33 | |
| RanpacModel Category=PTM-based, Backbone=ViT-B/16-IN1K, Throughput=154.64 (100%)2026.04 | 71.66 | 60.17 | |
| EASE + C-Flat TurboModel Category=PTM-based, Backbone=ViT-B/16-IN1K, Throughput=102.74 (61.6%)2026.04 | 64.96 | 52.61 | |
| EASE + C-FlatModel Category=PTM-based, Backbone=ViT-B/16-IN1K, Throughput=44.25 (26.5%)2026.04 | 64.89 | 52.47 | |
| L2P + C-Flat TurboModel Category=PTM-based, Backbone=ViT-B/16-IN1K, Throughput=65.50 (59.4%)2026.04 | 64.64 | 52.55 | |
| L2P + C-FlatModel Category=PTM-based, Backbone=ViT-B/16-IN1K, Throughput=28.63 (30.0%)2026.04 | 64.53 | 52.47 | |
| EASEModel Category=PTM-based, Backbone=ViT-B/16-IN1K, Throughput=166.67 (100%)2026.04 | 64.38 | 52.02 | |
| L2PModel Category=PTM-based, Backbone=ViT-B/16-IN1K, Throughput=110.29 (100%)2026.04 | 64.18 | 52.1 | |
| MEMO + C-Flat TurboModel Category=Typical, Backbone=ViT-B/16-IN1K, Throughput=91.91 (68.0%)2026.04 | 57.52 | 40.62 | |
| MEMO + C-FlatModel Category=Typical, Backbone=ViT-B/16-IN1K, Throughput=46.11 (34.1%)2026.04 | 56.5 | 39.45 | |
| MEMOModel Category=Typical, Backbone=ViT-B/16-IN1K, Throughput=135.14 (100%)2026.04 | 56.22 | 38.32 | |
| iCaRL + C-Flat TurboModel Category=Typical, Backbone=ViT-B/16-IN1K, Throughput=45.89 (62.6%)2026.04 | 50.49 | 29.3 | |
| iCaRL + C-FlatModel Category=Typical, Backbone=ViT-B/16-IN1K, Throughput=19.72 (26.9%)2026.04 | 49.59 | 29.03 | |
| iCaRLModel Category=Typical, Backbone=ViT-B/16-IN1K, Throughput=73.35 (100%)2026.04 | 48.06 | 28.2 |