Huge forgetting: wrong finetuning?

[caesar753]

Good morning,

I have been using avalanche on my dataset for a while (10k images divided into 66 classes) using a pre-trained resnet50 as I would like to implement new instances and, above all, new classes scenarios, however the trainings performed so far have been rather discouraging.
I used various strategies (EWC, LWF, GEM, SI) in addition to the naive, dividing the dataset into 11 experiences of 6 classes each but, unfortunately, I have noticed a huge forgetting between one experience and another with any strategy , to the point that the accuracy of previous experiences often reaches 0.
I also carried out tests on a much smaller dataset (only 6 classes), dividing it into 3 experiences of 2 classes each, and also in this case the result is the same as that obtained with the complete dataset.
Below some data from logs are attached

[caesar753]

SMALL DATASET: about 600 images divided into 6 classes

LWF strategy

-- >> Start of training phase << --
-- Starting training on experience 0 (Task 0) from train stream --
Epoch 0 ended.
	DiskUsage_Epoch/train_phase/train_stream/Task000 = 343435.1191
	DiskUsage_MB/train_phase/train_stream/Task000 = 343435.1191
	Loss_Epoch/train_phase/train_stream/Task000 = 1.7905
	Loss_MB/train_phase/train_stream/Task000 = 1.7905
	RunningTime_Epoch/train_phase/train_stream/Task000 = 55.0214
	Time_Epoch/train_phase/train_stream/Task000 = 55.0214
	Top1_Acc_Epoch/train_phase/train_stream/Task000 = 0.1529
	Top1_Acc_MB/train_phase/train_stream/Task000 = 0.1529
Epoch 4 ended.
	DiskUsage_Epoch/train_phase/train_stream/Task000 = 343437.0098
	DiskUsage_MB/train_phase/train_stream/Task000 = 343437.0098
	Loss_Epoch/train_phase/train_stream/Task000 = 1.3567
	Loss_MB/train_phase/train_stream/Task000 = 1.3567
	RunningTime_Epoch/train_phase/train_stream/Task000 = 57.1457
	Time_Epoch/train_phase/train_stream/Task000 = 57.1457
	Top1_Acc_Epoch/train_phase/train_stream/Task000 = 0.5669
	Top1_Acc_MB/train_phase/train_stream/Task000 = 0.5669
-- >> End of training phase << --
-- >> Start of eval phase << --
-- Starting eval on experience 0 (Task 0) from test stream --
> Eval on experience 0 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 2795.9000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 343437.6104
	Loss_Exp/eval_phase/test_stream/Task000/Exp000 = 1.1915
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000 = 0.6624
-- Starting eval on experience 1 (Task 0) from test stream --
> Eval on experience 1 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 2361.8000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 343437.9746
	Loss_Exp/eval_phase/test_stream/Task000/Exp001 = 2.1960
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp001 = 0.0000
-- Starting eval on experience 2 (Task 0) from test stream --
> Eval on experience 2 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 2574.7000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 343438.3389
	Loss_Exp/eval_phase/test_stream/Task000/Exp002 = 2.3791
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp002 = 0.0000
-- >> End of eval phase << --
	ConfusionMatrix_Stream/eval_phase/test_stream = 
tensor([[  0,  16,   0,  41,   0,   0],
        [  0,  42,   0,  33,   0,   0],
        [  0,  65,   0, 102,   0,   0],
        [  0,  20,   0,  62,   0,   0],
        [  0,  11,   0,  47,   0,   0],
        [  0,   7,   0,  28,   0,   0]])
	DiskUsage_Stream/eval_phase/test_stream/Task000 = 343438.3389
	Loss_Stream/eval_phase/test_stream/Task000 = 1.9413
	StreamForgetting/eval_phase/test_stream = 0.0000
	Top1_Acc_Stream/eval_phase/test_stream/Task000 = 0.2194

-- >> Start of training phase << --
-- Starting training on experience 1 (Task 0) from train stream --
[...]
Epoch 4 ended.
	DiskUsage_Epoch/train_phase/train_stream/Task000 = 343441.1777
	DiskUsage_MB/train_phase/train_stream/Task000 = 343441.1777
	Loss_Epoch/train_phase/train_stream/Task000 = 1.6041
	Loss_MB/train_phase/train_stream/Task000 = 1.6041
	RunningTime_Epoch/train_phase/train_stream/Task000 = 63.8121
	Time_Epoch/train_phase/train_stream/Task000 = 63.8121
	Top1_Acc_Epoch/train_phase/train_stream/Task000 = 0.2522
	Top1_Acc_MB/train_phase/train_stream/Task000 = 0.2522
-- >> End of training phase << --
-- >> Start of eval phase << --
-- Starting eval on experience 0 (Task 0) from test stream --
> Eval on experience 0 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 2544.6000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 343441.7783
	ExperienceForgetting/eval_phase/test_stream/Task000/Exp000 = 0.6306
	Loss_Exp/eval_phase/test_stream/Task000/Exp000 = 1.6791
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000 = 0.0318
-- Starting eval on experience 1 (Task 0) from test stream --
> Eval on experience 1 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 2375.5000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 343442.2109
	Loss_Exp/eval_phase/test_stream/Task000/Exp001 = 1.4413
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp001 = 0.5565
-- Starting eval on experience 2 (Task 0) from test stream --
> Eval on experience 2 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 2625.0000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 343442.5752
	Loss_Exp/eval_phase/test_stream/Task000/Exp002 = 2.5095
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp002 = 0.0000
-- >> End of eval phase << --
	ConfusionMatrix_Stream/eval_phase/test_stream = 
tensor([[29,  2,  0,  0, 26,  0],
        [49,  5,  0,  0, 21,  0],
        [83,  8,  0,  0, 76,  0],
        [25,  1,  0,  0, 56,  0],
        [21,  2,  0,  0, 35,  0],
        [ 7,  0,  0,  0, 28,  0]])
	DiskUsage_Stream/eval_phase/test_stream/Task000 = 343442.5752
	Loss_Stream/eval_phase/test_stream/Task000 = 1.9753
	StreamForgetting/eval_phase/test_stream = 0.6306
	Top1_Acc_Stream/eval_phase/test_stream/Task000 = 0.1456

-- Starting training on experience 2 (Task 0) from train stream --
[...]
Epoch 4 ended.
	DiskUsage_Epoch/train_phase/train_stream/Task000 = 343445.3789
	DiskUsage_MB/train_phase/train_stream/Task000 = 343445.3789
	Loss_Epoch/train_phase/train_stream/Task000 = 1.5599
	Loss_MB/train_phase/train_stream/Task000 = 1.5599
	RunningTime_Epoch/train_phase/train_stream/Task000 = 95.2316
	Time_Epoch/train_phase/train_stream/Task000 = 95.2316
	Top1_Acc_Epoch/train_phase/train_stream/Task000 = 0.6386
	Top1_Acc_MB/train_phase/train_stream/Task000 = 0.6386
-- >> End of training phase << --
-- >> Start of eval phase << --
-- Starting eval on experience 0 (Task 0) from test stream --
> Eval on experience 0 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 2795.5000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 343445.9795
	ExperienceForgetting/eval_phase/test_stream/Task000/Exp000 = 0.6624
	Loss_Exp/eval_phase/test_stream/Task000/Exp000 = 2.0240
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000 = 0.0000
-- Starting eval on experience 1 (Task 0) from test stream --
> Eval on experience 1 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 2355.6000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 343446.4121
	ExperienceForgetting/eval_phase/test_stream/Task000/Exp001 = 0.5565
	Loss_Exp/eval_phase/test_stream/Task000/Exp001 = 1.8956
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp001 = 0.0000
-- Starting eval on experience 2 (Task 0) from test stream --
> Eval on experience 2 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 2569.4000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 343446.8447
	Loss_Exp/eval_phase/test_stream/Task000/Exp002 = 1.2384
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp002 = 0.8267
-- >> End of eval phase << --
	ConfusionMatrix_Stream/eval_phase/test_stream = 
tensor([[  0,   0,  57,   0,   0,   0],
        [  0,   0,  75,   0,   0,   0],
        [  0,   0, 167,   0,   0,   0],
        [  0,   0,  82,   0,   0,   0],
        [  0,   0,  58,   0,   0,   0],
        [  0,   0,  35,   0,   0,   0]])
	DiskUsage_Stream/eval_phase/test_stream/Task000 = 343446.8447
	Loss_Stream/eval_phase/test_stream/Task000 = 1.6581
	StreamForgetting/eval_phase/test_stream = 0.6095
	Top1_Acc_Stream/eval_phase/test_stream/Task000 = 0.3523

[vlomonaco]

Hi @caesar753, regularization strategies are known to suffer significant forgetting in class-inc scenarios. Those you mentioned are also not necessarily easy to parametrize. You can take a look at those baselines in case your data is similar: https://github.com/ContinualAI/continual-learning-baselines

I suggest to start with replay-based methods for better results at least to validate your setting / data!

[caesar753]

ok, thanks!

[caesar753]

with replay strategies things works a lot better, this are the results on the small dataset:

6 classes in 3 experiences, 5 epochs

REPLAY STRATEGY

-- Starting eval on experience 0 (Task 0) from test stream --
> Eval on experience 0 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 2294.8000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 343517.1816
	ExperienceForgetting/eval_phase/test_stream/Task000/Exp000 = 0.3333
	Loss_Exp/eval_phase/test_stream/Task000/Exp000 = 1.6105
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000 = 0.2903
-- Starting eval on experience 1 (Task 0) from test stream --
> Eval on experience 1 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 2572.9000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 343517.6143
	ExperienceForgetting/eval_phase/test_stream/Task000/Exp001 = -0.2603
	Loss_Exp/eval_phase/test_stream/Task000/Exp001 = 1.6456
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp001 = 0.4835
-- Starting eval on experience 2 (Task 0) from test stream --
> Eval on experience 2 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 2572.3000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 343518.0479
	Loss_Exp/eval_phase/test_stream/Task000/Exp002 = 1.9888
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp002 = 0.0000
-- >> End of eval phase << --
	ConfusionMatrix_Stream/eval_phase/test_stream = 
tensor([[  0,   0,  43,   0,  14,   0],
        [  0,   0,  57,   0,  18,   0],
        [  0,   0, 117,   0,  50,   0],
        [  0,   0,  52,   0,  30,   0],
        [  0,   0,  31,   0,  27,   0],
        [  0,   0,  28,   0,   7,   0]])
	DiskUsage_Stream/eval_phase/test_stream/Task000 = 343518.0479
	Loss_Stream/eval_phase/test_stream/Task000 = 1.7394
	StreamForgetting/eval_phase/test_stream = 0.0365
	Top1_Acc_Stream/eval_phase/test_stream/Task000 = 0.3038

GDUMB STRATEGY

-- Starting eval on experience 0 (Task 0) from test stream --
> Eval on experience 0 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 2670.5000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 343539.9014
	ExperienceForgetting/eval_phase/test_stream/Task000/Exp000 = 0.4071
	Loss_Exp/eval_phase/test_stream/Task000/Exp000 = 1.8223
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000 = 0.1786
-- Starting eval on experience 1 (Task 0) from test stream --
> Eval on experience 1 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 2791.2000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 343540.3340
	ExperienceForgetting/eval_phase/test_stream/Task000/Exp001 = 0.1630
	Loss_Exp/eval_phase/test_stream/Task000/Exp001 = 1.7982
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp001 = 0.0978
-- Starting eval on experience 2 (Task 0) from test stream --
> Eval on experience 2 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 2782.0000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 343540.7666
	Loss_Exp/eval_phase/test_stream/Task000/Exp002 = 1.7432
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp002 = 0.2727
-- >> End of eval phase << --
	ConfusionMatrix_Stream/eval_phase/test_stream = 
tensor([[  4,  32,   3,  11,   0,   7],
        [  2,  53,  10,   7,   2,   1],
        [ 10, 100,  13,  12,   5,  27],
        [ 18,  24,   8,  24,   0,   8],
        [  4,  16,   8,  23,   1,   6],
        [  2,  11,   1,  16,   0,   5]])
	DiskUsage_Stream/eval_phase/test_stream/Task000 = 343540.7666
	Loss_Stream/eval_phase/test_stream/Task000 = 1.7772
	StreamForgetting/eval_phase/test_stream = 0.2851
	Top1_Acc_Stream/eval_phase/test_stream/Task000 = 0.2110

[caesar753]

SMALL DATASET: about 600 images divided into 6 classes

EWC strategy

-- >> Start of training phase << --
-- Starting training on experience 0 (Task 0) from train stream --
[...]
Epoch 4 ended.
	DiskUsage_Epoch/train_phase/train_stream/Task000 = 343449.6836
	DiskUsage_MB/train_phase/train_stream/Task000 = 343449.6836
	Loss_Epoch/train_phase/train_stream/Task000 = 1.2766
	Loss_MB/train_phase/train_stream/Task000 = 1.2766
	RunningTime_Epoch/train_phase/train_stream/Task000 = 57.8619
	Time_Epoch/train_phase/train_stream/Task000 = 57.8619
	Top1_Acc_Epoch/train_phase/train_stream/Task000 = 0.5860
	Top1_Acc_MB/train_phase/train_stream/Task000 = 0.5860
-- >> End of training phase << --
-- >> Start of eval phase << --
-- Starting eval on experience 0 (Task 0) from test stream --
> Eval on experience 0 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 2532.3000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 343450.2842
	Loss_Exp/eval_phase/test_stream/Task000/Exp000 = 1.1780
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000 = 0.6752
-- Starting eval on experience 1 (Task 0) from test stream --
> Eval on experience 1 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 2360.8000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 343450.6484
	Loss_Exp/eval_phase/test_stream/Task000/Exp001 = 2.3891
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp001 = 0.0000
-- Starting eval on experience 2 (Task 0) from test stream --
> Eval on experience 2 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 2568.0000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 343451.0127
	Loss_Exp/eval_phase/test_stream/Task000/Exp002 = 2.1988
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp002 = 0.0000
-- >> End of eval phase << --
	ConfusionMatrix_Stream/eval_phase/test_stream = 
tensor([[  0,  44,   0,  13,   0,   0],
        [  0,  70,   0,   5,   0,   0],
        [  0, 115,   0,  52,   0,   0],
        [  0,  46,   0,  36,   0,   0],
        [  0,  39,   0,  19,   0,   0],
        [  0,  27,   0,   8,   0,   0]])
	DiskUsage_Stream/eval_phase/test_stream/Task000 = 343451.0127
	Loss_Stream/eval_phase/test_stream/Task000 = 1.9069
	StreamForgetting/eval_phase/test_stream = 0.0000
	Top1_Acc_Stream/eval_phase/test_stream/Task000 = 0.2236

-- Starting training on experience 1 (Task 0) from train stream --
[...]
Epoch 4 ended.
	DiskUsage_Epoch/train_phase/train_stream/Task000 = 343453.8516
	DiskUsage_MB/train_phase/train_stream/Task000 = 343453.8516
	Loss_Epoch/train_phase/train_stream/Task000 = 1.7080
	Loss_MB/train_phase/train_stream/Task000 = 1.7080
	RunningTime_Epoch/train_phase/train_stream/Task000 = 51.4358
	Time_Epoch/train_phase/train_stream/Task000 = 51.4358
	Top1_Acc_Epoch/train_phase/train_stream/Task000 = 0.0696
	Top1_Acc_MB/train_phase/train_stream/Task000 = 0.0696
-- >> End of training phase << --
-- >> Start of eval phase << --
-- Starting eval on experience 0 (Task 0) from test stream --
> Eval on experience 0 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 2690.2000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 343454.4521
	ExperienceForgetting/eval_phase/test_stream/Task000/Exp000 = 0.5159
	Loss_Exp/eval_phase/test_stream/Task000/Exp000 = 1.6327
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000 = 0.1592
-- Starting eval on experience 1 (Task 0) from test stream --
> Eval on experience 1 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 2365.2000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 343454.8848
	Loss_Exp/eval_phase/test_stream/Task000/Exp001 = 1.5281
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp001 = 0.5043
-- Starting eval on experience 2 (Task 0) from test stream --
> Eval on experience 2 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 2562.0000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 343455.2490
	Loss_Exp/eval_phase/test_stream/Task000/Exp002 = 2.3490
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp002 = 0.0000
-- >> End of eval phase << --
	ConfusionMatrix_Stream/eval_phase/test_stream = 
tensor([[25, 14,  0,  0, 18,  0],
        [23, 25,  0,  1, 26,  0],
        [62, 45,  0,  2, 58,  0],
        [17, 13,  0,  0, 52,  0],
        [14,  8,  0,  3, 33,  0],
        [ 7,  6,  0,  0, 22,  0]])
	DiskUsage_Stream/eval_phase/test_stream/Task000 = 343455.2490
	Loss_Stream/eval_phase/test_stream/Task000 = 1.9126
	StreamForgetting/eval_phase/test_stream = 0.5159
	Top1_Acc_Stream/eval_phase/test_stream/Task000 = 0.1751

-- Starting training on experience 2 (Task 0) from train stream --
[...]

Epoch 4 ended.
	DiskUsage_Epoch/train_phase/train_stream/Task000 = 343458.0527
	DiskUsage_MB/train_phase/train_stream/Task000 = 343458.0527
	Loss_Epoch/train_phase/train_stream/Task000 = 1.5562
	Loss_MB/train_phase/train_stream/Task000 = 1.5562
	RunningTime_Epoch/train_phase/train_stream/Task000 = 74.8870
	Time_Epoch/train_phase/train_stream/Task000 = 74.8870
	Top1_Acc_Epoch/train_phase/train_stream/Task000 = 0.6733
	Top1_Acc_MB/train_phase/train_stream/Task000 = 0.6733
-- >> End of training phase << --
-- >> Start of eval phase << --
-- Starting eval on experience 0 (Task 0) from test stream --
> Eval on experience 0 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 2540.4000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp000 = 343458.6533
	ExperienceForgetting/eval_phase/test_stream/Task000/Exp000 = 0.6752
	Loss_Exp/eval_phase/test_stream/Task000/Exp000 = 1.9552
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp000 = 0.0000
-- Starting eval on experience 1 (Task 0) from test stream --
> Eval on experience 1 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 2356.8000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp001 = 343459.0859
	ExperienceForgetting/eval_phase/test_stream/Task000/Exp001 = 0.5043
	Loss_Exp/eval_phase/test_stream/Task000/Exp001 = 1.9595
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp001 = 0.0000
-- Starting eval on experience 2 (Task 0) from test stream --
> Eval on experience 2 (Task 0) from test stream ended.
	CPUUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 2563.7000
	DiskUsage_Exp/eval_phase/test_stream/Task000/Exp002 = 343459.5186
	Loss_Exp/eval_phase/test_stream/Task000/Exp002 = 1.2308
	Top1_Acc_Exp/eval_phase/test_stream/Task000/Exp002 = 0.8267
-- >> End of eval phase << --
	ConfusionMatrix_Stream/eval_phase/test_stream = 
tensor([[  0,   0,  57,   0,   0,   0],
        [  0,   0,  75,   0,   0,   0],
        [  0,   0, 167,   0,   0,   0],
        [  0,   0,  82,   0,   0,   0],
        [  0,   0,  58,   0,   0,   0],
        [  0,   0,  35,   0,   0,   0]])
	DiskUsage_Stream/eval_phase/test_stream/Task000 = 343459.5186
	Loss_Stream/eval_phase/test_stream/Task000 = 1.6476
	StreamForgetting/eval_phase/test_stream = 0.5898
	Top1_Acc_Stream/eval_phase/test_stream/Task000 = 0.3523