List your company
ResearchFormerly WizWand

Class Incremental Learning on ImageNet-1000 (incremental)

66.95Avg Incremental AccuracyPODNet (CNN)
43.404449.517255.6361.7428Apr 28, 2020Jun 18, 2020Aug 8, 2020Sep 29, 2020Nov 19, 2020Jan 9, 2021Mar 2, 2021
Updated 8mo ago

Evaluation results

MethodLinks
PODNet (CNN)Steps=5, New classes per step=100, Inference classifier=CNN2020.0466.95
LUCIR + DDER=5, T=5, Backbone=ResNet-182021.0366.18
UCIR (CNN)*Steps=5, New classes per step=100, Inference classifier=CNN, Source=Reported from Hou et al. [14]2020.0464.34
PODNet (CNN)Steps=10, New classes per step=50, Inference classifier=CNN2020.0464.13
MnemonicsR=20, T=10, Backbone=ResNet-182021.0363.01
LUCIRR=5, T=5, Backbone=ResNet-182021.0362.61
PODNet + DDER=10, T=10, Backbone=ResNet-182021.0362.09
UCIR (NME)*Steps=5, New classes per step=100, Inference classifier=NME, Source=Reported from Hou et al. [14]2020.0461.56
UCIR (CNN)*Steps=10, New classes per step=50, Inference classifier=CNN, Source=Reported from Hou et al. [14]2020.0461.28
UCIR (NME)*Steps=10, New classes per step=50, Inference classifier=NME, Source=Reported from Hou et al. [14]2020.0459.92
PODNetR=10, T=10, Backbone=ResNet-182021.0359.56
iCaRL*Steps=5, New classes per step=100, Source=Reported from Hou et al. [14]2020.0451.36
iCaRL*Steps=10, New classes per step=50, Source=Reported from Hou et al. [14]2020.0446.72
BiCSteps=5, New classes per step=1002020.0445.72
BiCSteps=10, New classes per step=502020.0444.31