About this document

EdgeNeXt: Efficient CNN-Transformer Model by 王大哥 is a document available to read on EtoBox.

EdgeNeXt is a new efficient hybrid CNN-Transformer architecture proposed for mobile vision applications. It introduces a split depth-wise transpose attention encoder that splits input tensors into channel groups and uses depth-wise convolution and self-attention across channels to implicitly increase receptive field and encode multi-scale features. Experiments show EdgeNeXt outperforms SOTA methods with lower compute requirements, achieving 71.2% top-1 accuracy on ImageNet-1K with 1.3M parameters and lower

Author

王大哥

Language

EN