Trait Encoder 

pub trait Encoder<B: Backend> {
    type Input;

    // Required methods
    fn encode(&self, input: &Self::Input) -> Representation<B>;
    fn embed_dim(&self) -> usize;
}

Trait for JEPA encoders.

An encoder maps raw input to a Representation with shape [batch, seq_len, embed_dim]. Concrete implementations include:

• jepa_vision::VitEncoder — Vision Transformer for images
• jepa_vision::VitVideoEncoder — Vision Transformer for video

Type parameters

• B — burn backend (e.g. NdArray, Wgpu, Tch)

Associated types

• Input — the raw input type this encoder accepts. For vision encoders this is typically a Tensor<B, 4> (images) or Tensor<B, 5> (video). Higher-level wrappers may accept Representation<B> so that levels in a hierarchy can chain.

Required Associated Types

type Input

The type of input this encoder accepts.

Required Methods

fn encode(&self, input: &Self::Input) -> Representation<B>

Encode input into a representation.

Arguments

• input - The raw input to encode

Returns

A Representation with shape [batch, seq_len, embed_dim]

fn embed_dim(&self) -> usize

Get the output embedding dimension.

Implementors