check.py

import argparse
import os

import numpy as np


PERCENTILES = [0, 0.1, 1, 5, 25, 50, 75, 95, 99, 99.9, 100]


def _as_float_array(values):
    return np.asarray(values, dtype=np.float64)


def _format_float(value):
    if value is None or not np.isfinite(value):
        return str(value)
    return f"{value:.8g}"


def _field_stats(values):
    arr = _as_float_array(values).reshape(-1)
    finite_mask = np.isfinite(arr)
    finite = arr[finite_mask]

    stats = {
        "count": int(arr.size),
        "finite": int(finite.size),
        "nan": int(np.isnan(arr).sum()),
        "pos_inf": int(np.isposinf(arr).sum()),
        "neg_inf": int(np.isneginf(arr).sum()),
        "zero": int((arr == 0).sum()),
        "neg": int((arr < 0).sum()),
        "pos": int((arr > 0).sum()),
    }

    if finite.size == 0:
        stats.update({
            "min": None,
            "max": None,
            "mean": None,
            "std": None,
            "abs_max": None,
            "percentiles": None,
        })
        return stats

    stats.update({
        "min": float(np.min(finite)),
        "max": float(np.max(finite)),
        "mean": float(np.mean(finite)),
        "std": float(np.std(finite)),
        "abs_max": float(np.max(np.abs(finite))),
        "percentiles": np.percentile(finite, PERCENTILES),
    })
    return stats


def _print_field_stats(name, values, top_extreme=0):
    stats = _field_stats(values)
    count = max(stats["count"], 1)
    finite_ratio = 100.0 * stats["finite"] / count
    zero_ratio = 100.0 * stats["zero"] / count

    print(f"\n[{name}]")
    print(
        "  count={count} finite={finite} ({finite_ratio:.2f}%) "
        "nan={nan} +inf={pos_inf} -inf={neg_inf}".format(
            finite_ratio=finite_ratio, **stats
        )
    )
    print(
        "  min={min} max={max} mean={mean} std={std} abs_max={abs_max}".format(
            min=_format_float(stats["min"]),
            max=_format_float(stats["max"]),
            mean=_format_float(stats["mean"]),
            std=_format_float(stats["std"]),
            abs_max=_format_float(stats["abs_max"]),
        )
    )
    print(
        "  zero={zero} ({zero_ratio:.2f}%) neg={neg} pos={pos}".format(
            zero_ratio=zero_ratio, **stats
        )
    )

    if stats["percentiles"] is not None:
        pairs = [
            f"p{p:g}={_format_float(v)}"
            for p, v in zip(PERCENTILES, stats["percentiles"])
        ]
        print("  " + " ".join(pairs))

    if top_extreme > 0:
        arr = _as_float_array(values).reshape(-1)
        finite_idx = np.flatnonzero(np.isfinite(arr))
        if finite_idx.size > 0:
            finite_vals = arr[finite_idx]
            k = min(top_extreme, finite_vals.size)
            order = np.argsort(np.abs(finite_vals))[-k:][::-1]
            items = [
                f"idx={int(finite_idx[i])}:value={_format_float(finite_vals[i])}"
                for i in order
            ]
            print("  top_abs: " + ", ".join(items))


def _existing_fields(names, prefix):
    return sorted(
        [name for name in names if name.startswith(prefix)],
        key=lambda x: int(x[len(prefix):]) if x[len(prefix):].isdigit() else x,
    )


def _stack_fields(vertex, fields):
    if not fields:
        return None
    return np.stack([vertex[name] for name in fields], axis=1).astype(np.float64)


def _print_vector_stats(label, arr):
    if arr is None:
        return
    finite_rows = np.isfinite(arr).all(axis=1)
    print(f"\n[{label} vector]")
    print(f"  shape={arr.shape} finite_rows={int(finite_rows.sum())}/{arr.shape[0]}")

    norms = np.linalg.norm(arr, axis=1)
    _print_field_stats(f"{label}_norm", norms)

    if arr.shape[1] == 3 and label == "scale_log":
        scale = np.exp(np.clip(arr, -80, 80))
        _print_field_stats("scale_actual_min_axis", scale.min(axis=1))
        _print_field_stats("scale_actual_max_axis", scale.max(axis=1))
        aspect = scale.max(axis=1) / np.maximum(scale.min(axis=1), 1e-12)
        _print_field_stats("scale_aspect_ratio", aspect)

    if arr.shape[1] == 4 and label == "rotation":
        norm = np.linalg.norm(arr, axis=1)
        bad = np.abs(norm - 1.0) > 1e-2
        print(f"  rotation_norm_not_1(>|1e-2|)={int(bad.sum())}/{arr.shape[0]}")


def inspect_ply(path, top_extreme=0, only_bad=False):
    try:
        from plyfile import PlyData
    except ImportError as exc:
        raise ImportError(
            "Missing dependency 'plyfile'. Install it with: pip install plyfile"
        ) from exc

    if not os.path.exists(path):
        raise FileNotFoundError(path)

    ply = PlyData.read(path)
    if "vertex" not in ply:
        raise ValueError("PLY does not contain a vertex element")

    vertex = ply["vertex"]
    names = list(vertex.data.dtype.names)
    n_points = len(vertex)

    print(f"file: {os.path.abspath(path)}")
    print(f"points: {n_points}")
    print(f"fields ({len(names)}): {', '.join(names)}")

    for name in names:
        values = vertex[name]
        stats = _field_stats(values)
        if only_bad and stats["nan"] == 0 and stats["pos_inf"] == 0 and stats["neg_inf"] == 0:
            continue
        _print_field_stats(name, values, top_extreme=top_extreme)

    scale_fields = [name for name in ["scale_0", "scale_1", "scale_2"] if name in names]
    rot_fields = [name for name in ["rot_0", "rot_1", "rot_2", "rot_3"] if name in names]
    pos_fields = [name for name in ["x", "y", "z"] if name in names]
    dc_fields = [name for name in ["f_dc_0", "f_dc_1", "f_dc_2"] if name in names]
    sh_fields = _existing_fields(names, "f_rest_")

    _print_vector_stats("position", _stack_fields(vertex, pos_fields) if len(pos_fields) == 3 else None)
    _print_vector_stats("scale_log", _stack_fields(vertex, scale_fields) if len(scale_fields) == 3 else None)
    _print_vector_stats("rotation", _stack_fields(vertex, rot_fields) if len(rot_fields) == 4 else None)
    _print_vector_stats("dc", _stack_fields(vertex, dc_fields) if len(dc_fields) == 3 else None)

    if sh_fields:
        sh = _stack_fields(vertex, sh_fields)
        _print_vector_stats("sh_rest", sh)
        _print_field_stats("sh_rest_all_values", sh.reshape(-1), top_extreme=top_extreme)

    if "opacity" in names:
        opacity_logit = _as_float_array(vertex["opacity"])
        opacity_sigmoid = 1.0 / (1.0 + np.exp(-np.clip(opacity_logit, -80, 80)))
        _print_field_stats("opacity_sigmoid", opacity_sigmoid)

    if "filter_3D" in names:
        unique = np.unique(np.asarray(vertex["filter_3D"]))
        print(f"\n[filter_3D unique]")
        print(f"  unique_count={len(unique)}")
        if len(unique) <= 20:
            print("  values=" + ", ".join(_format_float(float(v)) for v in unique))
        else:
            print("  first20=" + ", ".join(_format_float(float(v)) for v in unique[:20]))


def parse_args():
    parser = argparse.ArgumentParser(
        description="Inspect numeric distributions of all vertex fields in a PLY file."
    )
    parser.add_argument("ply_path", help="Path to the PLY file to inspect.")
    parser.add_argument(
        "--top_extreme",
        type=int,
        default=0,
        help="Print top-k absolute extreme values per field with flat indices.",
    )
    parser.add_argument(
        "--only_bad",
        action="store_true",
        help="Only print fields containing NaN or Inf. Combined diagnostics are still printed.",
    )
    return parser.parse_args()


if __name__ == "__main__":
    args = parse_args()
    inspect_ply(args.ply_path, top_extreme=args.top_extreme, only_bad=args.only_bad)