EfficientViT: Lightweight Multi-Scale Attention for On-Device Semantic Segmentation
demo
EfficientViT is a new family of vision models for efficient high-resolution dense prediction. The core building block of EfficientViT is a new lightweight multi-scale linear attention module that achieves global receptive field and multi-scale learning with only hardware-efficient operations.
image-20230919102753047
image-20230919102314552
image-20230919102354273
image-20230919104023799
本文分享自 iResearch666 微信公众号,前往查看
如有侵权,请联系 cloudcommunity@tencent.com 删除。
本文参与 腾讯云自媒体同步曝光计划 ,欢迎热爱写作的你一起参与!