DeiT

GPTKB entity

Statements (50)
Predicate Object
gptkbp:instanceOf Vision Transformer model
gptkbp:architecture gptkb:transformation
gptkbp:author gptkb:Armand_Joulin
gptkb:Edouard_Grave
gptkb:Hugo_Touvron
gptkb:Hervé_Jégou
gptkb:Gabriel_Synnaeve
gptkb:Matthieu_Cord
Alexandre Sablayrolles
gptkbp:citation Over 2000 (as of 2024)
gptkbp:competitor gptkb:ResNet
gptkb:ViT
gptkbp:developedBy gptkb:Facebook_AI_Research
gptkbp:distillationToken Yes
gptkbp:fullName gptkb:Data-efficient_Image_Transformer
gptkbp:hasVariant DeiT-Base
DeiT-Large
DeiT-Small
DeiT-Tiny
https://www.w3.org/2000/01/rdf-schema#label DeiT
gptkbp:input gptkb:illustrator
gptkbp:introducedIn 2020
gptkbp:license Apache 2.0
gptkbp:mainActivity Image classification
gptkbp:notableFor Data efficiency
Efficient training on smaller datasets
Knowledge distillation with a distillation token
No need for large-scale pretraining
gptkbp:notablePublication Training data-efficient image transformers & distillation through attention
https://arxiv.org/abs/2012.12877
gptkbp:openSource Yes
gptkbp:output Class label
gptkbp:parameter 22 million (DeiT-B)
5.7 million (DeiT-T)
86 million (DeiT-L)
gptkbp:platform gptkb:PyTorch
gptkbp:pretrainedWeightsAvailable Yes
gptkbp:publishedIn gptkb:ICML_2021
gptkbp:relatedTo gptkb:ImageNet
gptkb:Vision_Transformer
Knowledge distillation
gptkbp:repository https://github.com/facebookresearch/deit
gptkbp:resolution 224x224 pixels
gptkbp:trainer gptkb:ImageNet
gptkbp:usedIn Computer vision research
Fine-tuning for downstream tasks
Transfer learning
gptkbp:usesDistillation Yes
gptkbp:bfsParent gptkb:Vision_Transformer
gptkbp:bfsLayer 6