Skip to contents

Extract embeddings from a pretrained SimCLR encoder

Source: R/foundation_simclr.R
foundation_simclr_embed.Rd

Returns the backbone features (before the projection head) — these are the reusable vectors to feed into downstream DSM models. If you wanted the contrastive-projection vectors instead, pass projection = TRUE.

Usage

foundation_simclr_embed(object, patches, projection = FALSE)

Arguments

object

A edaphos_foundation_simclr.

patches

Array (N, C, H, W).

projection

Logical; return L2-normalised projection-head outputs instead of backbone features.

Value

Numeric matrix N x D with D = feature_dim (or proj_dim).