|
gptkbp:instanceOf
|
Convolutional Neural Network Model
|
|
gptkbp:activatedBy
|
gptkb:GELU
|
|
gptkbp:architecture
|
gptkb:convolutional_neural_network
|
|
gptkbp:author
|
gptkb:Zhuang_Liu
gptkb:Trevor_Darrell
gptkb:Saining_Xie
gptkb:Chao-Yuan_Wu
gptkb:Christoph_Feichtenhofer
gptkb:Hanzi_Mao
|
|
gptkbp:benchmarkPerformance
|
Top-1 accuracy 83.5% on ImageNet-1K
|
|
gptkbp:developedBy
|
gptkb:Facebook_AI_Research
|
|
gptkbp:family
|
gptkb:ConvNeXt
|
|
https://www.w3.org/2000/01/rdf-schema#label
|
ConvNeXt-Large
|
|
gptkbp:inspiredBy
|
gptkb:Vision_Transformer
|
|
gptkbp:introducedIn
|
2022
|
|
gptkbp:normalization
|
gptkb:LayerNorm
|
|
gptkbp:notableFeature
|
Depthwise Separable Convolutions
Competitive with Vision Transformers
Fewer activation functions
Large kernel sizes
Layer scaling
Modernized ConvNet design
No bottleneck structure
No squeeze-and-excitation blocks
Stochastic depth
|
|
gptkbp:notablePublication
|
A ConvNet for the 2020s
https://arxiv.org/abs/2201.03545
|
|
gptkbp:openSource
|
Yes
|
|
gptkbp:parameter
|
196 million
|
|
gptkbp:platform
|
gptkb:TensorFlow
gptkb:PyTorch
|
|
gptkbp:resolution
|
224x224
384x384
|
|
gptkbp:trainer
|
ImageNet-1K
ImageNet-22K
|
|
gptkbp:usedFor
|
Transfer Learning
Image Classification
|
|
gptkbp:bfsParent
|
gptkb:ConvNeXt
|
|
gptkbp:bfsLayer
|
8
|