miniImageNet

E899067

few-shot learning benchmark image dataset meta-learning benchmark

miniImageNet is a widely used few-shot learning benchmark dataset derived from ImageNet, consisting of small, labeled images across many classes for evaluating meta-learning and one-shot learning algorithms.

Try in SPARQL Jump to: Statements Referenced by

Statements (46)

Predicate	Object
instanceOf	few-shot learning benchmark ⓘ image dataset ⓘ meta-learning benchmark ⓘ
accessMethod	constructed from ImageNet image IDs ⓘ
benchmarkRole	standard baseline for few-shot learning ⓘ
commonlyUsedWithShotCounts	1-shot ⓘ 5-shot ⓘ
commonlyUsedWithWayCounts	20-way ⓘ 5-way ⓘ
comparedWith	CIFAR-FS NERFINISHED ⓘ FC100 NERFINISHED ⓘ tieredImageNet NERFINISHED ⓘ
dataFormat	image files with class labels ⓘ
derivedFrom	ImageNet NERFINISHED ⓘ
evaluationProtocol	episodic training and testing ⓘ
hasBenchmarkProperty	fixed train/validation/test class splits ⓘ supports reproducible few-shot experiments ⓘ
hasClassBalance	approximately balanced across classes ⓘ
hasClassCount	100 ⓘ
hasDataType	natural images ⓘ
hasDomain	general object recognition ⓘ
hasFewShotSetting	N-way K-shot classification ⓘ
hasGranularity	object category level ⓘ
hasInputShape	3x84x84 (channels-first) in many implementations ⓘ
hasLabelType	image class labels ⓘ
hasLicenseSource	ImageNet license ⓘ
hasModality	RGB images ⓘ
hasNumberOfClasses	100 ⓘ
hasTask	image classification ⓘ
hasTestSplitClassCount	20 ⓘ
hasTotalImageCountApprox	60000 ⓘ
hasTrainSplitClassCount	64 ⓘ
hasTypicalImageResolution	84x84 pixels ⓘ
hasTypicalTrainImagesPerClass	600 ⓘ
hasValidationSplitClassCount	16 ⓘ
introducedFor	benchmarking meta-learning algorithms ⓘ
isSubsetOf	ImageNet ILSVRC-2012 dataset NERFINISHED ⓘ
isWidelyUsedSince	around 2016 ⓘ
popularizedBy	Matching Networks for One Shot Learning NERFINISHED ⓘ Model-Agnostic Meta-Learning (MAML) NERFINISHED ⓘ Prototypical Networks for Few-shot Learning NERFINISHED ⓘ
usedFor	few-shot learning evaluation ⓘ meta-learning evaluation ⓘ one-shot learning evaluation ⓘ
usedIn	computer vision research ⓘ machine learning research ⓘ

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

Matching Networks for One Shot Learning → datasetUsed → miniImageNet ⓘ