{
  "task": "seg",
  "experiment": "num_points",
  "metric": "point accuracy",
  "baseline_accuracy": 0.9003974068071313,
  "q1_q2_reported_accuracy": 0.9003974068071313,
  "conditions": [
    {
      "condition": "10000 pts",
      "value": 10000,
      "num_points": 10000,
      "accuracy": 0.9003974068071313,
      "delta_pp": 0.0,
      "visualized": [
        {
          "object_id": 309,
          "accuracy": 0.9463
        },
        {
          "object_id": 551,
          "accuracy": 0.9118
        },
        {
          "object_id": 588,
          "accuracy": 0.9711
        }
      ]
    },
    {
      "condition": "5000 pts",
      "value": 5000,
      "num_points": 5000,
      "accuracy": 0.9011523500810372,
      "delta_pp": 0.07549432739059059,
      "visualized": [
        {
          "object_id": 309,
          "accuracy": 0.9438
        },
        {
          "object_id": 551,
          "accuracy": 0.9096
        },
        {
          "object_id": 588,
          "accuracy": 0.9716
        }
      ]
    },
    {
      "condition": "2500 pts",
      "value": 2500,
      "num_points": 2500,
      "accuracy": 0.900807131280389,
      "delta_pp": 0.0409724473257711,
      "visualized": [
        {
          "object_id": 309,
          "accuracy": 0.9428
        },
        {
          "object_id": 551,
          "accuracy": 0.9012
        },
        {
          "object_id": 588,
          "accuracy": 0.974
        }
      ]
    },
    {
      "condition": "1250 pts",
      "value": 1250,
      "num_points": 1250,
      "accuracy": 0.8966794165316045,
      "delta_pp": -0.37179902755267724,
      "visualized": [
        {
          "object_id": 309,
          "accuracy": 0.9424
        },
        {
          "object_id": 551,
          "accuracy": 0.9032
        },
        {
          "object_id": 588,
          "accuracy": 0.9664
        }
      ]
    }
  ],
  "settings": {
    "checkpoint": "./checkpoints/seg/best_model.pt",
    "checkpoint_epoch": 248,
    "checkpoint_val_acc": 0.8914093541202672,
    "checkpoint_config": {
      "task": "seg",
      "num_cls_class": 3,
      "num_seg_class": 6,
      "num_points": 10000,
      "num_epochs": 250,
      "batch_size": 32,
      "num_workers": 0,
      "lr": 0.001,
      "val_frac": 0.1,
      "seed": 0,
      "device": "cuda",
      "exp_name": "exp",
      "main_dir": "./data/",
      "checkpoint_dir": "./checkpoints/seg",
      "log_dir": "./logs",
      "checkpoint_every": 10,
      "load_checkpoint": "",
      "overwrite": false
    },
    "data_dir": "./data/",
    "num_points": 10000,
    "seed": 0,
    "batch_size": 16,
    "device": "cuda",
    "point_sampling": "per-object random permutation of all available points from numpy default_rng(seed); the first num_points indices are used, so smaller subsets are nested in larger ones",
    "rotation_axis": "x",
    "rotation_angles_deg": [
      0,
      15,
      30,
      60,
      90
    ],
    "point_fractions": [
      1.0,
      0.5,
      0.25,
      0.125
    ],
    "rotation_convention": "points are row vectors p and are replaced by R p, with R the right-handed rotation about the +x axis; the data is Y-up (objects stand along +Y). Rotation is applied after point selection, so point identity and labels are unchanged",
    "num_points_convention": "the first k indices of the same per-object permutation used for the baseline (nested subsets, sampled without replacement); seg labels use the same indices",
    "visualized_object_ids": [
      309,
      551,
      588
    ]
  }
}