HyperAI초신경

홈 뉴스 최신 연구 논문 튜토리얼 데이터셋 백과사전 SOTA LLM 모델 GPU 랭킹 컨퍼런스

한국어

HyperAI초신경

소수 샷 이미지 분류

Few-Shot 이미지 분류는 몇 개의 라벨된 샘플(일반적으로 6개 미만)을 사용하여 새로운 이미지를 분류하는 컴퓨터 비전 작업입니다. 이 작업의 목표는 모델이 최소한의 감독과 데이터 요구 사항으로 새로운 카테고리를 신속하게 인식하고 분류할 수 있도록 하는 것입니다. 이를 통해 제한된 데이터 조건에서도 모델의 일반화 능력을 향상시킵니다. 이 기술은 특히 데이터 수집이 어려운 경우나 비용이 많이 드는 상황에서 실질적인 가치를 지니고 있습니다.

Mini-Imagenet 5-way (1-shot)

PEMnE-BMS* (transductive)

Mini-Imagenet 5-way (5-shot)

CAML [Laion-2b]

Tiered ImageNet 5-way (5-shot)

CAML [Laion-2b]

Tiered ImageNet 5-way (1-shot)

CIFAR-FS 5-way (5-shot)

PT+MAP+SF+SOT (transductive)

CIFAR-FS 5-way (1-shot)

PT+MAP+SF+SOT (transductive)

CUB 200 5-way 1-shot

PT+MAP+SF+BPA (transductive)

CUB 200 5-way 5-shot

PT+MAP+SF+SOT (transductive)

FC100 5-way (1-shot)

FC100 5-way (5-shot)

OMNIGLOT - 1-Shot, 20-way

OMNIGLOT - 5-Shot, 20-way

OMNIGLOT - 1-Shot, 5-way

Mini-ImageNet - 1-Shot Learning

OMNIGLOT - 5-Shot, 5-way

Mini-Imagenet 10-way (1-shot)

Transductive CNAPS + FETI

Mini-Imagenet 10-way (5-shot)

Transductive CNAPS + FETI

Meta-Dataset Rank

Tiered ImageNet 10-way (1-shot)

Transductive CNAPS + FETI

Tiered ImageNet 10-way (5-shot)

Transductive CNAPS + FETI

Dirichlet Mini-Imagenet (5-way, 1-shot)

Dirichlet Mini-Imagenet (5-way, 5-shot)

Mini-ImageNet-CUB 5-way (1-shot)

Dirichlet Tiered-Imagenet (5-way, 1-shot)

Dirichlet Tiered-Imagenet (5-way, 5-shot)

Dirichlet CUB-200 (5-way, 1-shot)

Dirichlet CUB-200 (5-way, 5-shot)

ImageNet - 1-shot

ViT-MoE-15B (Every-2)

ImageNet - 5-shot

ViT-MoE-15B (Every-2)

ImageNet-FS (2-shot, novel)

ImageNet-FS (5-shot, all)

KGTN-ens (ResNet-50, h+g, max)

Mini-ImageNet-CUB 5-way (5-shot)

Human (Amateur)

ImageNet - 10-shot

ViT-MoE-15B (Every-2)

ImageNet-FS (1-shot, novel)

Mini-Imagenet 20-way (1-shot)

Mini-Imagenet 20-way (5-shot)

Stanford Dogs 5-way (5-shot)

Stanford Cars 5-way (1-shot)

Stanford Cars 5-way (5-shot)

CUB-200-2011 - 0-Shot

Word CNN-RNN (DS-SJE Embedding)

ImageNet - 0-Shot

CLIP (ViT B/32)

Mini-Imagenet 5-way (10-shot)

CUB 200 50-way (0-shot)

Prototypical Networks

Caltech-256 5-way (1-shot)

CUB-200 - 0-Shot Learning

ImageNet-FS (5-shot, novel)

ORBIT Clutter Video Evaluation

Stanford Dogs 5-way (1-shot)

CIFAR100 5-way (1-shot)

ImageNet (1-shot)

ImageNet-FS (10-shot, novel)

ImageNet-FS (1-shot, all)

ImageNet-FS (2-shot, all)

ImageNet-FS (10-shot, all)

KGTN (ResNet-50)

Mini-ImageNet to CUB - 5 shot learning

OMNIGLOT-EMNIST 5-way (1-shot)

OMNIGLOT-EMNIST 5-way (5-shot)

ORBIT Clean Video Evaluation

SimpleCNAPs + LITE

Synthesised Classifier

Synthesised Classifier

Caltech-256 5-way (5-shot)

MergedNet-Concat

CIFAR-FS - 1-Shot Learning

CIFAR-FS - 5-Shot Learning

CUB-200-2011 5-way (1-shot)

CUB-200-2011 5-way (5-shot)

EASY 3xResNet12 (transductive)

FC100 5-way (10-shot)

Fewshot-CIFAR100 - 1-Shot Learning

Fewshot-CIFAR100 - 5-Shot Learning

Flowers-102 - 0-Shot

Word CNN-RNN (DS-SJE Embedding)

iNaturalist (227-way multi-shot)

iNaturalist 2018 - 1-shot

iNaturalist 2018 - 5-shot

iNaturalist 2018 - 10-shot

mini-ImageNet - 100-Way

miniImagenet → CUB (5-way 1-shot)

miniImagenet → CUB (5-way 5-shot)

OMNIGLOT - 1-Shot, 423 way

OMNIGLOT - 1-Shot, 1000 way

OMNIGLOT - 5-Shot, 423 way

OMNIGLOT - 5-Shot, 1000 way

Oxford 102 Flower