# How does the embeddings work in vision transformer from paper?

I get the part from the paper where the image is split into P say 16x16 (smaller images) patches and then you have to Flatten the 3-D (16,16,3) patch to pass it into a Linear layer to get what they call "Liner Projection". After passing from the Linear layer, the patches will be vectors but with some "meaning" to them.

Can someone please explain how the two types of embeddings are working?

I visited this implementation on github, looked at the code too and looked like a maze to me.

If someone could just explain how these embeddings are working in laymen's terms, I'll look at the code again and understand.

• Which paper are you referring to? Please, provide the link to the paper and clarify the exact section where they are talking about these "embeddings". – nbro Jan 30 at 2:04
• People who know Vision Transformer will get that. But I'll add. Thanks. – Deshwal Jan 30 at 6:52