Reproducible configs and checkpoints
This folder contains:
Reproducible config of Table.1 in the paper of LibFewShot
.
Reproducible config of Table.2 in the paper of LibFewShot
.
Reproduction results on miniImageNet
(Results may different from the paper. Here are up-to-date results with checkpoints and configs.)
Method
Embed.
5-way 1-shot
5-way 5-shot
Reported
Ours
Reported
Ours
Baseline
Conv64F
42.11
-
62.53
-
ResNet18
51.75
51.18 ± 0.34
74.27
74.06 ± 0.28
Baseline++
Conv64F
48.24
-
66.43
-
ResNet18
51.87
-
75.68
-
RFS-simple
ResNet12
62.02 ± 0.63
62.80 ± 0.52
79.64 ± 0.44
79.57± 0.39
RFS-distill
ResNet12
64.82
63.44
82.14
80.17
SKD-GEN0
ResNet12
65.93
66.40
83.15
83.06
SKD-GEN1
ResNet12
67.04
67.35
83.54
80.30
RENet
ResNet12
67.60 ± 0.44
66.83 ± 0.36
82.58 ± 0.30
82.13 ± 0.26
MAML
Conv32F
48.70
47.41
63.11
65.24
Versa
Conv64F†
53.40
51.92
67.37
66.26
R2D2
Conv64F
49.50
47.57
65.40
66.68
Conv64F‡
51.80
55.53
68.40
70.79
ANIL
Conv32F
46.70
48.44
61.50
64.35
BOIL
Conv64F
49.61 ± 0.16
**48.00 ± 0.36
66.45 ± 0.37
**64.39 ± 0.30
ResNet12**
-
**58.87 ± 0.38
71.30 ± 0.28
**72.88 ± 0.29
MTL
ResNet12
60.20
60.20
74.30
75.86
ProtoNet†
Conv64F
46.14
46.30
65.77
66.24
RelationNet
Conv64F
50.44
51.75
65.32
66.77
CovaMNet
Conv64F
51.19
53.36
67.65
68.17
DN4
Conv64F
51.24
51.95
71.02
71.42
ResNet12†
54.37
57.76
74.44
77.57
CAN
ResNet12
63.85
66.62
79.44
78.96
NegCos
ResNet12
63.85 ± 0.81
63.28 ± 0.36
81.57 ± 0.56
81.24 ± 0.26
The overview picture of the SOTAs
Method
Venue
Type
miniImageNet
tieredImageNet
1-shot
5-shot
1-shot
5-shot
Baseline
ICLR’19
Fine-tuning
44.90 ± 0.32
63.96 ± 0.30
48.20 ± 0.35
68.96 ± 0.33
Baseline++
ICML’19
Fine-tuning
48.86 ± 0.35
63.29 ± 0.30
55.94 ± 0.39
73.80 ± 0.32
RFS-simple
ECCV’20
Fine-tuning
47.97 ± 0.33
65.88 ± 0.30
52.21 ± 0.37
71.82 ± 0.32
RFS-distill
ECCV’20
Fine-tuning
-
-
-
-
SKD-GEN0
arXiv’20
Fine-tuning
48.14 ± 0.33
66.36 ± 0.29
51.78 ± 0.36
70.65 ± 0.32
SKD-GEN1
arXiv’20
Fine-tuning
-
-
-
-
NegCos
ECCV’20
Fine-tuning
47.34
65.97
51.21
71.57
RENet
ICCV’21
Fine-tuning
57.62 ± 0.36
74.14 ± 0.27
61.62 ± 0.40
76.74 ± 0.33
MAML
ICML’17
Meta
49.55 ± 0.37
64.92 ± 0.32
50.98 ± 0.43
67.12 ± 0.35
Versa
NeurIPS’18
Meta
52.75 ± 0.38
67.40 ± 0.31
52.28 ± 0.99
69.41 ± 0.37
R2D2
ICLR’19
Meta
51.19 ± 0.36
67.29 ± 0.31
52.18 ± 0.40
69.19 ± 0.36
LEO
ICLR’19
Meta
53.31 ± 0.37
67.47 ± 0.30
58.15 ± 0.40
74.21 ± 0.33
MTL
CVPR’19
Meta
40.97
57.12
42.36
64.87
ANIL
ICLR’20
Meta
48.01 ± 0.35
63.88 ± 0.32
49.05 ± 0.39
66.32 ± 0.34
BOIL
ICLR’21
Meta
47.92 ± 0.35
64.39 ± 0.30
50.04 ± 0.38
65.51 ± 0.34
ProtoNet
NeurIPS’17
Metric
47.05 ± 0.35
68.56 ± 0.16
46.11 ± 0.39
70.07 ± 0.34
RelationNet
CVPR’18
Metric
51.52 ± 0.37
66.76 ± 0.30
54.37 ± 0.44
71.93 ± 0.35
CovaMNet
AAAI’19
Metric
51.59 ± 0.36
67.65 ± 0.32
51.92 ± 0.40
69.76 ± 0.34
DN4
CVPR’19
Metric
54.47 ± 0.36
72.15 ± 0.29
56.07 ± 0.38
75.75 ± 0.31
CAN
NeurIPS’19
Metric
55.88 ± 0.38
70.98 ± 0.30
55.96 ± 0.42
70.52 ± 0.35
Method
Venue
Type
miniImageNet
tieredImageNet
1-shot
5-shot
1-shot
5-shot
Baseline
ICLR’19
Fine-tuning
56.39 ± 0.36
76.18 ± 0.27
-
-
Baseline++
ICML’19
Fine-tuning
56.75 ± 0.38
66.36 ± 0.29
65.95 ± 0.42
82.25 ± 0.31
RFS-simple
ECCV’20
Fine-tuning
61.65 ± 0.35
78.88 ± 0.25
70.55 ± 0.42
84.74 ± 0.29
RFS-distill
ECCV’20
Fine-tuning
-
-
-
-
SKD-GEN0
arXiv’20
Fine-tuning
66.40 ± 0.36
83.06 ± 0.24
-
-
SKD-GEN1
arXiv’20
Fine-tuning
67.35 ± 0.37
83.31 ± 0.24
-
-
NegCos
ECCV’20
Fine-tuning
60.60 ±
78.80 ±
70.15 ±
84.94 ±
RENet
ICCV’21
Fine-tuning
64.81 ± 0.37
79.90 ± 0.27
70.14 ± 0.43
82.70 ± 0.31
Versa
NeurIPS’18
Meta
55.71 ± 0.40
70.05 ± 0.31
57.14 ± 0.43
75.48 ± 0.34
R2D2
ICLR’19
Meta
59.52 ± 0.39
74.61 ± 0.30
65.07 ± 0.44
83.04 ± 0.30
LEO
ICLR’19
Meta
53.58 ± 0.39
68.24 ± 0.32
-
-
MTL
CVPR’19
Meta
-
-
-
-
ANIL
ICLR’20
Meta
52.77 ± 0.40
68.11 ± 0.34
55.65 ± 0.44
73.53 ± 0.35
BOIL
ICLR’21
Meta
58.87 ± 0.38
72.88 ± 0.29
64.66 ± 0.44
80.38 ± 0.31
ProtoNet
NeurIPS’17
Metric
58.61 ± 0.38
75.02 ± 0.28
62.93 ± 0.43
83.30 ± 0.29
RelationNet
CVPR’18
Metric
55.22 ± 0.39
69.25 ± 0.31
-
-
CovaMNet
AAAI’19
Metric
56.95 ± 0.37
71.41 ± 0.66
58.49 ± 0.42
76.34 ± 0.37
DN4
CVPR’19
Metric
58.68 ± 0.37
74.70 ± 0.28
64.41 ± 0.42
82.59 ± 0.30
CAN
NeurIPS’19
Metric
59.82 ± 0.38
76.54 ± 0.29
70.46 ± 0.43
84.50 ± 0.30
Method
Venue
Type
miniImageNet
tieredImageNet
1-shot
5-shot
1-shot
5-shot
Baseline
ICLR’19
Fine-tuning
54.11 ± 0.35
74.44 ± 0.29
64.65 ± 0.41
82.73 ± 0.29
Baseline++
ICML’19
Fine-tuning
-
-
-
-
RFS-simple
ECCV’20
Fine-tuning
61.65 ± 0.37
76.60 ± 0.28
69.14 ± 0.42
83.21 ± 0.31
RFS-distill
ECCV’20
Fine-tuning
-
-
-
-
SKD-GEN0
arXiv’20
Fine-tuning
66.18 ± 0.37
82.21 ± 0.24
70.00 ± 0.57
84.70 ± 0.41
SKD-GEN1
arXiv’20
Fine-tuning
66.70 ± 0.37
82.60 ± 0.24
NegCos
ECCV’20
Fine-tuning
60.99 ±
76.30 ±
68.36 ±
83.77 ±
RENet
ICCV’21
Fine-tuning
62.86 ± 0.37
-
71.53 ± 0.43
-
Versa
NeurIPS’18
Meta
55.08 ± 0.39
69.16 ± 0.30
57.30 ± 0.45
75.67 ± 0.38
R2D2
ICLR’19
Meta
58.36 ± 0.38
75.69 ± 0.29
64.73 ± 0.44
83.40 ± 0.31
LEO
ICLR’19
Meta
57.51 ± 0.39
69.33 ± 0.38
64.02 ± 0.44
78.89 ± 0.33
MTL
CVPR’19
Meta
-
-
-
-
ANIL
ICLR’20
Meta
52.96 ± 0.40
65.88 ± 0.33
55.81 ± 0.46
73.53 ± 0.37
BOIL
ICLR’21
Meta
57.85 ± 0.40
70.84 ± 0.29
60.85 ± 0.45
77.74 ± 0.38
ProtoNet
NeurIPS’17
Metric
58.48 ± 0.39
75.16 ± 0.29
63.50 ± 0.48
82.51 ± 0.30
RelationNet
CVPR’18
Metric
53.98 ± 0.37
71.27 ± 0.31
-
-
CovaMNet
AAAI’19
Metric
55.83 ± 0.39
70.97 ± 0.33
54.12 ± 0.47
73.51 ± 0.55
DN4
CVPR’19
Metric
57.92 ± 0.37
75.50 ± 0.28
64.83 ± 0.42
82.77 ± 0.30
CAN(!re-run,h2=11!)
NeurIPS’19
Metric
60.78 ± 0.40
75.05 ± 0.29
71.70 ± 0.43
84.61 ± 0.37