Task-Incremental Learning on Split-ImageNet 10 tasks
64.7Avg AccOGP
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OGPProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 64.7 | 16.2 | 50.5 | |
| ER (5%)Protocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 63.4 | 17.5 | 46.5 | |
| Adam-NSCLProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 62.1 | 19.1 | 41.6 | |
| TRGPProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 61.8 | 19.8 | 39.4 | |
| SGPProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 60.4 | 21.3 | 34.9 | |
| EWCProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 55.8 | 27.1 | 17.1 | |
| Sequential FTProtocol=Protocol A, Learning Setting=From-scratch, Backbone=ViT-B/16, Optimizer=Adam2026.04 | 51.2 | 32.7 | — |