List your company
ResearchFormerly WizWand

5-way Few-shot Classification on CUB (test)

82.9AccuracyFEAT
50.046458.575767.10575.6343Dec 10, 2018Dec 29, 2018Jan 17, 2019Feb 5, 2019Feb 24, 2019Mar 15, 2019Apr 3, 2019
Updated 8mo ago

Evaluation results

MethodLinks
FEATBackbone=ConvNet, Setup=Embedding Adaptation, Test trials=100002018.1282.9——
Hyperbolic ProtoNetEmbedding Net=4 Conv2019.0482.53——
DN4-DAEmbedding Net=4 Conv2019.0481.9——
ProtoNetBackbone=ConvNet, Setup=Instance Embedding, Test trials=100002018.1281.5——
GCNBackbone=ConvNet, Setup=Embedding Adaptation, Test trials=100002018.1280.26——
DEEPSETSBackbone=ConvNet, Setup=Embedding Adaptation, Test trials=100002018.1279.65——
Baseline++Embedding Net=4 Conv2019.0479.34——
MatchNetBackbone=ConvNet, Setup=Instance Embedding, Test trials=100002018.1279——
RelationNetEmbedding Net=4 Conv2019.0476.11——
RelationNetBackbone=ConvNet2018.1276.11——
MACOEmbedding Net=4 Conv2019.0474.96——
BILSTMBackbone=ConvNet, Setup=Embedding Adaptation, Test trials=100002018.1273.51——
MatchingNetEmbedding Net=4 Conv2019.0472.86——
MatchNetBackbone=ConvNet2018.1272.86——
MAMLEmbedding Net=4 Conv2019.0472.09——
MAMLBackbone=ConvNet2018.1272.09——
ProtoNetEmbedding Net=4 Conv2019.0470.77——
ProtoNetBackbone=ConvNet2018.1270.77——
FEATBackbone=ConvNet, Setup=Embedding Adaptation, Test trials=100002018.1268.87——
GCNBackbone=ConvNet, Setup=Embedding Adaptation, Test trials=100002018.1267.83——
MatchNetBackbone=ConvNet, Setup=Instance Embedding, Test trials=100002018.1267.73——
DEEPSETSBackbone=ConvNet, Setup=Embedding Adaptation, Test trials=100002018.1267.22——
Hyperbolic ProtoNetEmbedding Net=4 Conv2019.0464.02——
ProtoNetBackbone=ConvNet, Setup=Instance Embedding, Test trials=100002018.1263.72——
RelationNetEmbedding Net=4 Conv2019.0462.45——
RelationNetBackbone=ConvNet2018.1262.45——
BILSTMBackbone=ConvNet, Setup=Embedding Adaptation, Test trials=100002018.1262.05——
MatchingNetEmbedding Net=4 Conv2019.0461.16——
MatchNetBackbone=ConvNet2018.1261.16——
MACOEmbedding Net=4 Conv2019.0460.76——
Baseline++Embedding Net=4 Conv2019.0460.53——
MAMLEmbedding Net=4 Conv2019.0455.92——
MAMLBackbone=ConvNet2018.1255.92——
DN4-DAEmbedding Net=4 Conv2019.0453.15——
ProtoNetEmbedding Net=4 Conv2019.0451.31——
ProtoNetBackbone=ConvNet2018.1251.31——
Baseline2019.05—47.1264.16
BaselineSetting=Inductive, Backbone=ResNet-18, Input Size=224x2242020.03—65.5182.85
Baseline ++2019.05—60.5379.34
Baseline++Setting=Inductive, Backbone=ResNet-18, Input Size=224x2242020.03—67.0283.58
LRSetting=Inductive, Backbone=ResNet-12, Input Size=84x842020.03—76.1690.32
LR + ICISetting=Transductive, Backbone=ResNet-12, Input Size=84x842020.03—88.0692.53
LR + ICISetting=Semi-supervised, Backbone=ResNet-12, Input Size=84x84, Unlabeled Samples=15/152020.03—87.2892.18
LR + ICISetting=Semi-supervised, Backbone=ResNet-12, Input Size=84x84, Unlabeled Samples=30/502020.03—89.5892.48
LR + ICISetting=Semi-supervised, Backbone=ResNet-12, Input Size=84x84, Unlabeled Samples=80/802020.03—91.1192.98
MAML2019.05—55.9272.09
MAMLSetting=Inductive, Backbone=ResNet-18, Input Size=224x2242020.03—69.9682.7
MAML++Complexity=Low-End2019.05—62.1976.08
MAML++Complexity=High-End2019.05—67.4883.8
MAML++ + SCAComplexity=Low-End2019.05—66.1377.62
MAML++ + SCAComplexity=High-End2019.05—70.4685.63
Matching networks2019.05—61.1672.86
MatchingNetSetting=Inductive, Backbone=ResNet-18, Input Size=224x2242020.03—72.3683.64
Nearest neighbor (baseline)Features=ImageNet model features (trained on disjoint categories)2018.06—58.780.2
ProtoNetSetting=Inductive, Backbone=ResNet-18, Input Size=224x2242020.03—71.8887.42
Prototypical NetworksFeatures=ImageNet model features (trained on disjoint categories)2018.06—71.992.4
RelationNetSetting=Inductive, Backbone=ResNet-18, Input Size=224x2242020.03—67.5982.75
SVMSetting=Inductive, Backbone=ResNet-12, Input Size=84x842020.03—75.8489.26
SVM + ICISetting=Transductive, Backbone=ResNet-12, Input Size=84x842020.03—87.8792.38
SVM + ICISetting=Semi-supervised, Backbone=ResNet-12, Input Size=84x84, Unlabeled Samples=15/152020.03—86.8391.58
SVM + ICISetting=Semi-supervised, Backbone=ResNet-12, Input Size=84x84, Unlabeled Samples=30/502020.03—88.9492.14
TEAMSetting=Transductive, Backbone=ResNet-18, Input Size=224x2242020.03—80.1687.17
Δ-encoderFeatures=ImageNet model features (trained on disjoint categories)2018.06—82.292.6