Struct PromptEncoderOutput 

pub struct PromptEncoderOutput {
    pub sparse_embeddings: Vec<f32>,
    pub num_sparse_tokens: usize,
    pub dense_embeddings: Vec<f32>,
    pub image_pe: Vec<f32>,
}

Output of [PromptEncoder::forward] — fed straight into the mask decoder. All host-side Vec<f32>.

Fields

sparse_embeddings: Vec<f32>

[num_tokens, embed_dim] — concatenation of point and box embeddings. num_tokens = 0 for the “no prompt” case.

num_sparse_tokens: usize

dense_embeddings: Vec<f32>

[embed_dim, hw, hw] — dense pixel-wise embedding. Either the mask-downscaled signal or the broadcast no_mask_embed.

image_pe: Vec<f32>

[embed_dim, hw, hw] — image positional encoding (the dense PE fed into the mask decoder).