List your company

Class-Incremental Learning on Stanford Cars (T=10, test)

77.5Avg AccuracyRanPAC
36.10846.85457.668.346Jul 5, 2023Dec 25, 2023Jun 15, 2024Dec 5, 2024May 27, 2025Nov 16, 2025May 8, 2026

Evaluation results

MethodLinks
RanPACAlgorithm=12023.0777.562——
CPrompt2024.0376.81—66.7713.95
CODAprompt2024.0373.28—62.2415.08
RanPAC (No RPs)RP layer=false2023.0773———
ESN2024.0372.82—56.9113.5
L2P2024.0371.92—60.3913
TRMBackbone=CLIP ViT-B/16, Method Category=Model Merging2026.0570.4———
DualPrompt2024.0370.36—57.2716.31
LwFMemory=Theta(1)2025.0969.84———
CMMMemory=Theta(1)2025.0969.7———
CLG-CBMYear=2025, Backbone=CLIP ViT-B/16, Method Category=PEFT2026.0568.7———
CODAPromptYear=2023, Backbone=CLIP ViT-B/16, Method Category=PEFT2026.0567.4———
DualPromptYear=2023, Backbone=CLIP ViT-B/16, Method Category=PEFT2026.0567.1———
L2PYear=2022, Backbone=CLIP ViT-B/16, Method Category=PEFT2026.0566.7———
TIESYear=2023, Backbone=CLIP ViT-B/16, Method Category=Model Merging2026.0566.1———
RanPAC (No Phase 1)Phase 1=false2023.0766———
MagMaxYear=2024, Backbone=CLIP ViT-B/16, Method Category=Model Merging2026.0565.9———
RandMixMemory=Theta(N)2025.0965.62———
AvgMemory=Theta(N)2025.0965.59———
TIESMemory=Theta(N)2025.0964.9———
Model StockYear=2024, Backbone=CLIP ViT-B/16, Method Category=Model Merging2026.0564.5———
MaxAbsMemory=Theta(N)2025.0963.55———
RanPAC (No RPs or Phase 1)RP layer=false, Phase 1=false2023.0762.2———
LwFYear=2017, Backbone=CLIP ViT-B/16, Method Category=Conventional2026.0561.7———
EWCYear=2017, Backbone=CLIP ViT-B/16, Method Category=Conventional2026.0561.3———
EWCMemory=Theta(1)2025.0960.78———
Joint linear probeJoint training=true2023.0751.7———
CODA-Prompt2023.0743.2———
ADaM2023.0741.4———
RanPAC (NCM with Phase 1)Evaluation=NCM, Phase 1=true2023.0740.517——
DualPrompt2023.0740.1———
L2P2023.0738.2———
RanPAC (NCM only)Evaluation=NCM, Phase 1=false2023.0737.710——
UB2024.03——83.96—