data/eval/cii/manifest.json

{
  "layout_version": 1,
  "path_policy": "All image paths are relative to the repository root.",
  "description": "CII validation data for direct spatial-tool output imitation; separate from the public benchmark suite.",
  "datasets": {
    "locate": {
      "path": "data/eval/cii/locate.validation.200.verl_sft.parquet",
      "rows": 200,
      "visual_refs": 220,
      "unique_visual_refs": 220
    },
    "segment": {
      "path": "data/eval/cii/segment.validation.200.verl_sft.parquet",
      "rows": 200,
      "visual_refs": 215,
      "unique_visual_refs": 181
    },
    "depth": {
      "path": "data/eval/cii/depth.validation.200.verl_sft.parquet",
      "rows": 200,
      "visual_refs": 227,
      "unique_visual_refs": 222
    },
    "pose-object": {
      "path": "data/eval/cii/pose-object.validation.200.verl_sft.parquet",
      "rows": 200,
      "visual_refs": 244,
      "unique_visual_refs": 223
    },
    "pose-camera": {
      "path": "data/eval/cii/pose-camera.validation.200.verl_sft.parquet",
      "rows": 200,
      "visual_refs": 407,
      "unique_visual_refs": 390
    }
  },
  "total_rows": 1000,
  "total_visual_refs": 1313,
  "unique_visual_refs": 1181,
  "asset_root": "data/assets"
}