File size: 5,321 Bytes
68afb88
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
#!/usr/bin/env python3
"""
preprocess_step3_ptv3.py
=========================
NEST3D Pre-processing Step 3: Convert corrected PLY files to Pointcept format.

Input:  <data-dir>/sampleXXX/sampleXXX_corrected.ply
        <split-dir>/{train,val,test}.txt  (one sample ID per line)
Output: <out-dir>/train|val|test/sampleXXX/{coord.npy, color.npy, segment.npy}

Format:
  coord.npy   : (N, 3) float32 - XYZ coordinates, centered on the scene
                                  centroid and normalized to the unit
                                  sphere (max radius = 1)
  color.npy   : (N, 3) float32 - RGB colors scaled to [0, 1]
  segment.npy : (N,)   int32   - semantic labels: 0=grass, 1=tree, 2=nest,
                                  -1=ignore

Semantic labels (0/1/2) are preserved as-is from the corrected PLY files;
the ignore label is remapped from 255 (CloudCompare convention) to -1
(Pointcept convention).

The train/val/test split is defined by train.txt, val.txt, and test.txt
(one sample ID per line) and is not hardcoded here -- this script simply
reads and applies whatever split those files define. This keeps the split
definition in one authoritative place.

Usage:
  python preprocess_step3_ptv3.py \\
      --data-dir /path/to/reconstructions \\
      --split-dir /path/to/split/txt/files \\
      --out-dir /path/to/output

Author: NEST3D team
"""

import argparse
import numpy as np
from pathlib import Path
from plyfile import PlyData

# ── Helpers ───────────────────────────────────────────────────────────────────
def load_split(split_dir, split_name):
    path = split_dir / f"{split_name}.txt"
    with open(path) as f:
        samples = [line.strip() for line in f if line.strip()]
    return samples

def load_ply(path):
    ply = PlyData.read(str(path))
    v   = ply["vertex"]
    xyz = np.stack([v["x"], v["y"], v["z"]], axis=1).astype(np.float32)
    rgb = np.stack([v["red"], v["green"], v["blue"]], axis=1).astype(np.float32) / 255.0
    lbl = np.array(v["scalar_Classification"], dtype=np.float32).astype(np.int32)
    return xyz, rgb, lbl

def unit_sphere(xyz):
    centroid = xyz.mean(axis=0)
    xyz = xyz - centroid
    max_radius = np.sqrt((xyz**2).sum(axis=1)).max()
    if max_radius > 0:
        xyz = xyz / max_radius
    return xyz.astype(np.float32)

def remap_ignore(lbl):
    """Only remap 255 -> -1 (ignore). Keep 0, 1, 2 as-is."""
    out = lbl.copy()
    out[lbl == 255] = -1
    return out.astype(np.int32)

def save_sample(xyz, rgb, lbl, out_dir):
    out_dir.mkdir(parents=True, exist_ok=True)
    np.save(str(out_dir / "coord.npy"),   xyz)
    np.save(str(out_dir / "color.npy"),   rgb)
    np.save(str(out_dir / "segment.npy"), lbl)

# ── Main ──────────────────────────────────────────────────────────────────────
def main():
    parser = argparse.ArgumentParser(description="NEST3D Step 3: convert to Pointcept format")
    parser.add_argument("--data-dir", type=Path, default=Path("./reconstructions"),
                         help="Path to reconstructions/ containing sampleXXX_corrected.ply files")
    parser.add_argument("--split-dir", type=Path, default=Path("."),
                         help="Path to the directory containing train.txt, val.txt, test.txt")
    parser.add_argument("--out-dir", type=Path, default=Path("./nest"),
                         help="Output directory for the Pointcept-format data")
    args = parser.parse_args()

    recon_dir = args.data_dir
    out_dir   = args.out_dir

    all_splits = [(name, load_split(args.split_dir, name)) for name in ["train", "val", "test"]]

    for split, samples in all_splits:
        (out_dir / split).mkdir(parents=True, exist_ok=True)
        print(f"\n=== {split.upper()} ({len(samples)} samples) ===")
        for sample_id in samples:
            sample_out_dir = out_dir / split / sample_id
            ply_path = recon_dir / sample_id / f"{sample_id}_corrected.ply"

            if all((sample_out_dir / f).exists() for f in ["coord.npy","color.npy","segment.npy"]):
                print(f"  [SKIP] {sample_id}")
                continue

            if not ply_path.exists():
                print(f"  [MISSING] {sample_id}: {ply_path.name} not found")
                continue

            print(f"  [PROC] {sample_id} ...", end=" ", flush=True)
            xyz, rgb, lbl = load_ply(ply_path)
            lbl = remap_ignore(lbl)
            xyz = unit_sphere(xyz)

            n_total = len(lbl)
            n_grass = int((lbl==0).sum())
            n_tree  = int((lbl==1).sum())
            n_nest  = int((lbl==2).sum())
            n_ign   = int((lbl==-1).sum())

            save_sample(xyz, rgb, lbl, sample_out_dir)
            print(f"n={n_total:,} | grass={100*n_grass/n_total:.1f}% "
                  f"tree={100*n_tree/n_total:.1f}% nest={100*n_nest/n_total:.1f}% "
                  f"ignore={100*n_ign/n_total:.1f}%")

    print(f"\nDone! Output: {out_dir}")
    for split, samples in all_splits:
        print(f"  {split}: {len(samples)} samples")

if __name__ == "__main__":
    main()