configs/config_v2_vitl14.json

{
    "generic": {
        "seed": 13,
        "deterministic": true
    },
    "training": {},
    "data": {
        "image_shape": [420, 560],
        "shape_constraints": {
            "ratio_bounds": [0.66, 2.0],
            "pixels_bounds": [1400, 2400],
            "patch_size": 14
        }
    },
    "model": {
        "name": "UniDepthV2",
        "num_heads": 8,
        "expansion": 4,
        "pixel_decoder": {
            "hidden_dim": 512,
            "depths": [6, 0, 0],
            "dropout": 0.0
        },
        "pixel_encoder": {
            "name": "dinov2_vitl14",
            "pretrained": null,
            "use_norm": true,
            "stacking_fn": "last",
            "output_idx": [21,22,23,24]
        }
    }
}