{
  "backbone_name": "dinov3_base",
  "backbone_pool_size": 2,
  "backbone_patch_size": 16,
  "latent_downsampling_rate": 8,
  "max_uncertain_patches": 16,
  "encoder_num_layers": 12,
  "ae_decoder_num_layers": 4,
  "latent_decoder_num_layers": 12,
  "head_dim": 64,
  "num_attention_heads": 12,
  "intermediate_size": 3072,
  "latent_hidden_size": 32,
  "latent_expand_size": 2,
  "dropout": 0.1,
  "max_perception_tokens": 16
}