Generalized Few-Shot Learning research benchmarks

Benchmarks

Dataset NameSOTA methodMetricTrendResultsLast Updated

ImageNet 2012 (Novel classes)

Dynamic FSL + TRAML

79.4Top-5 Accuracy

Feb 26, 2026

ImageNet All classes 2012

Dynamic FSL + TRAML

80.2Top-5 Accuracy

Feb 26, 2026

AWA2

APN+TF-VAEGAN

96Accuracy

Feb 26, 2026

AWA Two-Level distributions (test)

DRAGON

83.3Harmonic Mean Accuracy (Acc_H)

Feb 26, 2026

CUB Two-Level distributions (test)

DRAGON

69.9Harmonic Mean Accuracy (Acc_H)

Feb 26, 2026

Two-Level SUN distributions (test)

DRAGON

48.2Harmonic Mean Acc_H

Feb 26, 2026

ImageNet base and novel classes 1K

APN(w2v)+f-VAEGAN-D2

84.3Top-5 Accuracy

Feb 26, 2026

SUN

APN+TF-VAEGAN

46.7Accuracy

Feb 26, 2026

CUB

APN+f-VAEGAN-D2

80.9Accuracy

Feb 26, 2026

miniImageNet GFSL

MVCN

78.7Accuracy (Novel)

Feb 27, 2026

tiered-ImageNet

MVCN

83.06Novel Accuracy

Feb 26, 2026

Mini-ImageNet + 20 novel ImageNet classes (test)

Few-Shot Learning with Global Class Representations

30.36Accuracy (Setting A)

Feb 27, 2026