config.json

{
  "data": {
    "augmentations": {
      "affine_p": 0.0,
      "blur_p": 0.5,
      "cut_p": 0.0,
      "flip_p": 0.5,
      "gamma_p": 1.0,
      "grayscale_p": 0.2,
      "invert_p": 0.0,
      "jitter_p": 1.0,
      "noise_pad": 1.0,
      "random_blur": 2.0,
      "random_gamma": 0.5,
      "random_jitter": 0.5,
      "random_rotation": 0.0,
      "random_scale": 4.0,
      "random_shear": 0.0,
      "random_translate_x": 0.04,
      "random_translate_y": 0.01,
      "rotation_p": 0.0,
      "scale_p": 0.0,
      "shape_mult": 14,
      "test_context": 1.0,
      "translate_p": 0.0
    },
    "crop": "garg",
    "data_root": "datasets",
    "image_shape": [
      518,
      518
    ],
    "mini": 1.0,
    "normalization": "imagenet",
    "num_frames": 1,
    "pair": 1,
    "resize_method": "contextcrop",
    "sampling": {
      "KITTI": 1.0
    },
    "shape_constraints": {
      "height_min": 15,
      "pixels_max": 600000.0,
      "pixels_min": 200000.0,
      "ratio_bounds": [
        0.5,
        2.5
      ],
      "sample": true,
      "shape_mult": 14,
      "width_min": 15
    },
    "train_datasets": [
      "KITTI"
    ],
    "val_datasets": [
      "KITTI"
    ]
  },
  "eps": 1e-06,
  "generic": {
    "deterministic": true,
    "name_page": "ufish",
    "seed": 42
  },
  "model": {
    "camera": {
      "augment": true,
      "tau": 50000,
      "weak_ratio": 0.9
    },
    "expansion": 4,
    "layer_scale": 0.0001,
    "name": "UniK3D",
    "num_heads": 8,
    "num_steps": 100000,
    "pixel_decoder": {
      "depths": [
        2,
        2,
        2
      ],
      "detach": 0.1,
      "dropout": 0.0,
      "hidden_dim": 256,
      "kernel_size": 3,
      "name": "Decoder",
      "num_prompt_blocks": 1,
      "out_dim": 32,
      "use_norm": false
    },
    "pixel_encoder": {
      "cls_token_embed_dims": [
        384,
        384,
        384,
        384,
        384,
        384,
        384,
        384,
        384,
        384,
        384,
        384
      ],
      "depths": [
        3,
        6,
        9,
        12
      ],
      "embed_dim": 384,
      "embed_dims": [
        384,
        384,
        384,
        384,
        384,
        384,
        384,
        384,
        384,
        384,
        384,
        384
      ],
      "freeze_norm": true,
      "frozen_stages": 0,
      "lr": 3e-06,
      "name": "dinov2_vits14",
      "num_register_tokens": 0,
      "output_idx": [
        3,
        6,
        9,
        12
      ],
      "pretrained": null,
      "stacking_fn": "last",
      "use_norm": true,
      "wd": 0.1
    }
  },
  "training": {
    "batch_size": 8,
    "clipping": 1.0,
    "cycle_beta": true,
    "drop_path": 0.0,
    "ema": 0.9995,
    "f16": "f16",
    "ld": 1.0,
    "losses": {
      "camera": {
        "alpha": 1.0,
        "dims": [
          1,
          2
        ],
        "fn": "l1",
        "gamma": 1.0,
        "input_fn": "linear",
        "name": "PolarRegression",
        "output_fn": "sqrt",
        "polar_asym": 0.7,
        "polar_weight": 3.0,
        "weight": 1.0
      },
      "confidence": {
        "input_fn": "log",
        "name": "Confidence",
        "output_fn": "sqrt",
        "weight": 0.1
      },
      "scale": {
        "alpha": 1.0,
        "fn": "l1",
        "gamma": 1.0,
        "input_fn": "log",
        "name": "Scale",
        "output_fn": "sqrt",
        "weight": 1.0
      }
    },
    "lr": 5e-05,
    "lr_final": 1e-06,
    "lr_warmup": 1.0,
    "n_iters": 250000,
    "nsteps_accumulation_gradient": 4,
    "validation_interval": 2500,
    "warmup_iters": 75000,
    "wd": 0.1,
    "wd_final": 0.1
  }
}