ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders
{, }
Paper: https://arxiv.org/pdf/2301.00808.pdf
Code:
Embedded Files
Motivation, Objectives and Related Works
Motivation, Objectives and Related Works
Motivation
Motivation
Objectives
Objectives
Fully Convolutional Masked Autoencoder Framework.
Global Response Normalization (GRN): inter-channel feature competition.
Related Works
Related Works
Model
Model
Idea
Idea
Model
Model
Encoder
Encoder
ConvNeXt.
Sparse data perspective: 2D sparse array of pixels.
Decoder
Decoder
ConvNeXt.
Fully Convolutional Masked Autoencoder
Fully Convolutional Masked Autoencoder
Mask Ratio: 60%.
Global Response Normalization
Global Response Normalization
Feature Collapse.
Feature Cosine Distance Analysis.
Steps:
Global Feature.
Feature Normalization.
Feature Cabliration.
Loss Function
Loss Function
Mean Squared Error (MSE).
Experimental Results
Experimental Results
Dataset
Dataset
ImageNet Classification.
COCO Detection.
ADE20K Segmentation.
Metrics
Metrics
Experimental Results
Experimental Results
Ablations
Ablations
References
References
n2 n0
θ
Page updated
Google Sites
Report abuse