pae-dinov3-large-d32/decoder/config.json

{
  "_class_name": "PAEDecoder",
  "decoder_patch_size": 16,
  "encoder_config_path": "facebook/dinov3-vitl16-pretrain-lvd1689m",
  "encoder_hidden_size": 1024,
  "encoder_input_size": 256,
  "encoder_num_heads": 16,
  "encoder_patch_size": 16,
  "latent_dim": 32,
  "latent_mean": [
    -1.140625,
    0.03759765625,
    -0.275390625,
    -0.34375,
    -0.265625,
    0.015869140625,
    0.1884765625,
    0.0087890625,
    1.625,
    -0.005279541015625,
    0.34765625,
    -1.03125,
    0.47265625,
    -1.4375,
    0.00604248046875,
    0.484375,
    -0.1328125,
    0.5234375,
    -0.11865234375,
    -0.37109375,
    0.474609375,
    -1.5390625,
    0.7734375,
    -0.087890625,
    0.039794921875,
    0.734375,
    -0.030517578125,
    -0.158203125,
    -3.078125,
    0.0869140625,
    0.138671875,
    -0.74609375
  ],
  "latent_multiplier": 1.0,
  "latent_std": [
    0.5390625,
    0.53515625,
    0.62109375,
    0.5546875,
    0.53125,
    0.44140625,
    0.53125,
    0.53125,
    0.55078125,
    0.58203125,
    0.74609375,
    0.48046875,
    0.49609375,
    0.5390625,
    0.59765625,
    0.5703125,
    0.55859375,
    0.66015625,
    0.55859375,
    0.50390625,
    0.55078125,
    0.490234375,
    0.7109375,
    0.55859375,
    0.53515625,
    0.486328125,
    0.5234375,
    0.515625,
    0.466796875,
    0.5390625,
    0.59375,
    0.51171875
  ]
}