Struct PatchEmbedding 

pub struct PatchEmbedding {
    pub projection_w: Tensor,
    pub projection_b: Tensor,
    pub image_size: usize,
    pub patch_size: usize,
    pub in_channels: usize,
    pub embed_dim: usize,
    pub num_patches: usize,
}

Patch embedding layer for Vision Transformer.

Splits an NHWC image into non-overlapping patches and projects each patch to an embedding vector via a linear projection (equivalent to Conv2d with kernel_size=patch_size, stride=patch_size).

Input: [batch, H, W, C] (NHWC) Output: [batch, num_patches, embed_dim]

Fields

projection_w: Tensor

Linear projection weight: [patch_size * patch_size * in_channels, embed_dim]

projection_b: Tensor

Linear projection bias: [embed_dim]

image_size: usize

patch_size: usize

in_channels: usize

embed_dim: usize

num_patches: usize

Implementations

impl PatchEmbedding

pub fn new( image_size: usize, patch_size: usize, in_channels: usize, embed_dim: usize, ) -> Result<Self, ModelError>

Creates a zero-initialized patch embedding.

pub fn forward(&self, input: &Tensor) -> Result<Tensor, ModelError>

Forward: [batch, H, W, C] -> [batch, num_patches, embed_dim].