Included models:
- autoencoder.safetensors : for sparse, clustered neuroimages
- proj_head_text_infonce.safetensors : contrastive text projection, from normalized specter2 embeddings
- proj_head_image_infonce.safetensors : contrastive image projection, from normalized specter2 embeddings
- proj_head_text_mse.safetensors : projects from normalized specter2 embeddings to latent autoencoder space, for text-to-brain generation
- masker.npz : used to resampled nifti images, masking, and flattening to ~28K dimensional vectors, the size needed for the autoencoder.
