config.json

{
  "generic": {
    "seed": 13
  },
  "training": {},
  "data": {
    "image_shape": [
      462,
      616
    ]
  },
  "model": {
    "name": "UniDepthV1",
    "num_heads": 8,
    "expansion": 4,
    "pixel_decoder": {
      "hidden_dim": 512,
      "depths": [
        3,
        2,
        1
      ],
      "dropout": 0.0
    },
    "pixel_encoder": {
      "name": "dinov2_vitl14",
      "pretrained": null,
      "patch_size": 14,
      "embed_dim": 1024,
      "embed_dims": [
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024,
        1024
      ],
      "depths": [
        5,
        12,
        18,
        24
      ]
    }
  }
}