{
  "updated": "2026-10-22",
  "scenes": [
    "Duo",
    "Flat",
    "G2"
  ],
  "conditions": [
    "CentralPinhole",
    "Contiguous30",
    "Random30",
    "Hole54",
    "Hole90"
  ],
  "tasks": [
    "Cup",
    "Tape",
    "Pen"
  ],
  "decisions_per_task": 15,
  "total_task_decisions": 45,
  "protocol": "Leave one scene out; fit task-specific thresholds on the other two scenes to maximize balanced accuracy, then sensitivity, then smallest oriented threshold; single-class training uses constant predictions. All-three is fitted separately.",
  "rows": [
    {
      "method": "PSNR / Test",
      "correct_by_task": {
        "Cup": 8,
        "Tape": 8,
        "Pen": 15
      },
      "total_correct": 31,
      "total_decisions": 45,
      "all_three_correct": 9,
      "all_three_decisions": 15
    },
    {
      "method": "SSIM / Test",
      "correct_by_task": {
        "Cup": 8,
        "Tape": 9,
        "Pen": 15
      },
      "total_correct": 32,
      "total_decisions": 45,
      "all_three_correct": 7,
      "all_three_decisions": 15
    },
    {
      "method": "LPIPS / Test",
      "correct_by_task": {
        "Cup": 8,
        "Tape": 5,
        "Pen": 15
      },
      "total_correct": 28,
      "total_decisions": 45,
      "all_three_correct": 10,
      "all_three_decisions": 15
    },
    {
      "method": "NVS-SQA / Test",
      "correct_by_task": {
        "Cup": 7,
        "Tape": 8,
        "Pen": 15
      },
      "total_correct": 30,
      "total_decisions": 45,
      "all_three_correct": 13,
      "all_three_decisions": 15
    },
    {
      "method": "NVS-SQA / OBS",
      "correct_by_task": {
        "Cup": 13,
        "Tape": 4,
        "Pen": 15
      },
      "total_correct": 32,
      "total_decisions": 45,
      "all_three_correct": 4,
      "all_three_decisions": 15
    },
    {
      "method": "Always reliable",
      "correct_by_task": {
        "Cup": 8,
        "Tape": 12,
        "Pen": 15
      },
      "total_correct": 35,
      "total_decisions": 45,
      "all_three_correct": 6,
      "all_three_decisions": 15
    }
  ],
  "housing_removal": {
    "images": 950,
    "resolution": [
      518,
      518
    ],
    "fovy_degrees": 90,
    "frames_per_condition": {
      "Flat": 19,
      "Duo": 38,
      "G2": 38
    },
    "method": "Use housing-present and housing-excluded paired renders to identify affected pixels. Replace only those pixels in archived inputs; preserve all other original pixels exactly.",
    "camera_poses_unchanged": true,
    "backgrounds_unchanged": true,
    "model_unchanged": true,
    "test_scores_unchanged": true
  },
  "interpretation": "Pen has no ranking-disagreement labels and its perfect accuracy reflects constant predictions, not failure detection."
}