dataset_info.json

{
  "name": "deva_libero",
  "views": [
    "agentview",
    "eye_in_hand"
  ],
  "tile": [
    [
      "agentview",
      "eye_in_hand"
    ]
  ],
  "video": {
    "native_hw": [
      432,
      768
    ],
    "panel_hw": [
      432,
      384
    ],
    "fps": 16
  },
  "affordance": {
    "layout": "per_view",
    "dtype": "float16",
    "normalization": "none",
    "native_hw": {
      "agentview": [
        128,
        128
      ],
      "eye_in_hand": [
        128,
        128
      ]
    },
    "view_size": [
      128,
      128
    ],
    "arrays": {
      "<view>": "float16 [T, Hv, Wv] - raw affordance, un-normalized; the loader resizes to view_size then per-frame max-normalizes"
    }
  },
  "depth": {
    "layout": "per_view",
    "dtype": "float16",
    "range": [
      0,
      1
    ],
    "normalization": "per_view_per_clip_at_export",
    "native_hw": {
      "agentview": [
        128,
        128
      ],
      "eye_in_hand": [
        128,
        128
      ]
    },
    "valid_mask": "derived from the tile grid (null cells are invalid); not stored",
    "arrays": {
      "<view>": "float16 [T, Hv, Wv] in [0,1], per-view per-clip normalized at export"
    },
    "source": "DepthAnything (vitl) disparity -> depth = 1/disparity; higher value = farther",
    "clip": {
      "clip_percentile": 98.0,
      "note": "far tail clipped at p98 before per-view min-max; ~2% of pixels saturate at 1.0"
    }
  }
}