Table 2.

Mean average loss $mAL$ and average forgetting $F$ of compared methods

Methods	$mAL\downarrow$ (TeWoR)a	$F\downarrow$ (TeWoR)	$mAL\downarrow$ (TeWR)a	$F\downarrow$ (TeWR)
Fine-tune	7.889	1.368	7.924	1.284
IID-Offline	3.470	—b	3.520	—
EWC	8.199	1.712	8.209	1.754
SI	7.657	1.344	7.686	1.382
BufferSize$\left|\mathcal{M}\right|=\mathbf{1}\mathit{K}$
ER	10.723	1.843	10.554	1.749
ER-RM	5.132	0.285	5.169	0.284
NI-WLc	5.630	0.175	5.543	0.175
NI-WL-RDc	5.255	0.215	5.243	0.208
NI-WL-RMc	5.100	0.198	5.089	0.192
NI-WL-RM-KDc	4.950	0.169	5.066	0.194
BufferSize$\left|\mathcal{M}\right|=\mathbf{5}\mathit{K}$
ER	5.094	0.406	5.147	0.416
ER-RM	5.077	0.096	5.143	0.092
NI-WL	5.144	0.091	5.197	0.076
NI-WL-RD	4.768	0.071	4.891	0.073
NI-WL-RM	4.855	0.064	4.887	0.068
NI-WL-RM-KD	4.490	0.062	4.511	0.063
BufferSize$\left|\mathcal{M}\right|=\mathbf{10}\mathit{K}$
ER	4.975	0.420	4.980	0.412
ER-RM	4.948	0.082	5.035	0.091
NI-WL	4.996	0.082	4.988	0.079
NI-WL-RD	4.810	0.058	4.853	0.066
NI-WL-RM	4.558	0.068	4.574	0.066
NI-WL-RM-KD	4.029	0.069	4.045	0.071

Bold underline, italic underline, and underline font highlights the first place, second place, and third place with same BufferSize, respectively.

^aTeWoR is short for test set without rotation, TeWR is short for test set with rotation.

^b“—” means not applicable.

^cNI-WL, NI-WL-RD, NI-WL-RM, and NI-WL-RM-KD are four variants of our proposed method.