config.json

{
  "adaptors_kwargs": {
    "flow": {
      "class": "FlowAdaptor",
      "kwargs": {
        "base_shape": [
          224,
          224
        ],
        "flow_mean": [
          0.0,
          0.0
        ],
        "flow_std": [
          25,
          25
        ],
        "name": "flow",
        "output_normalized_coordinate": false,
        "scale_strategy": "scale_both"
      }
    }
  },
  "detach_uncertainty_head": false,
  "encoder_kwargs": {
    "data_norm_type": "dinov2",
    "name": "dinov2_encoder",
    "patch_size": 14,
    "size": "large",
    "with_registers": false
  },
  "encoder_str": "dinov2",
  "feature_head_kwargs": {
    "dpt_feature": {
      "feature_dim": 256,
      "hooks": [
        0,
        1,
        2,
        3
      ],
      "input_feature_dims": [
        1024,
        768,
        768,
        768
      ],
      "layer_dims": [
        96,
        192,
        384,
        768
      ],
      "output_width_ratio": 1,
      "patch_size": 14,
      "pretrained_checkpoint_path": null,
      "use_bn": false
    },
    "dpt_processor": {
      "hidden_dims": [
        128,
        128
      ],
      "input_feature_dim": 256,
      "output_dim": 2,
      "pretrained_checkpoint_path": null
    }
  },
  "head_type": "dpt",
  "inference_resolution": [
    560,
    420
  ],
  "info_sharing_and_head_structure": "dual+single",
  "info_sharing_kwargs": {
    "attn_drop": 0.0,
    "custom_positional_encoding": null,
    "depth": 12,
    "dim": 768,
    "drop_path": 0.0,
    "gradient_checkpointing": false,
    "indices": [
      5,
      8
    ],
    "init_values": null,
    "input_embed_dim": 1024,
    "max_num_views": 2,
    "mlp_ratio": 4.0,
    "name": "global_attention",
    "norm_intermediate": true,
    "num_heads": 12,
    "pretrained_checkpoint_path": null,
    "proj_drop": 0.0,
    "qk_norm": false,
    "qkv_bias": true,
    "size": "base",
    "use_rand_idx_pe_for_non_reference_views": false
  },
  "info_sharing_str": "global_attention",
  "pretrained_backbone_checkpoint_path": null,
  "pretrained_checkpoint_path": null,
  "uncertainty_adaptors_kwargs": {
    "non_occluded_mask": {
      "class": "MaskAdaptor",
      "kwargs": {
        "name": "non_occluded_mask"
      }
    }
  },
  "uncertainty_head_kwargs": {
    "dpt_feature": {
      "feature_dim": 256,
      "hooks": [
        0,
        1,
        2,
        3
      ],
      "input_feature_dims": [
        1024,
        768,
        768,
        768
      ],
      "layer_dims": [
        96,
        192,
        384,
        768
      ],
      "output_width_ratio": 1,
      "patch_size": 14,
      "pretrained_checkpoint_path": null,
      "use_bn": false
    },
    "dpt_processor": {
      "hidden_dims": [
        128,
        128
      ],
      "input_feature_dim": 256,
      "output_dim": 1,
      "pretrained_checkpoint_path": null
    }
  },
  "uncertainty_head_type": "dpt"
}