No PatchEmbeding Issue - Githubissues

rayleizhu / BiFormer

[CVPR 2023] Official code release of our paper "BiFormer: Vision Transformer with Bi-Level Routing Attention"

https://arxiv.org/abs/2303.08810

MIT License

461 stars 36 forks source link

No PatchEmbeding Issue #51

Closed Flash-Alita closed 3 weeks ago

Flash-Alita commented 4 weeks ago

Hello, there: I have noticed that all biformer models take nchw inputs, including the old_legacy version. Can I ask you why you don't adopt the PatchEmbeding method? And what do you think about PatchEmbeding? WIll transformers preform better without PatchEmbeding? Do you have some conclusion? thx...