File size: 3,215 Bytes
ecb7b71 | 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 | """
CLI entry point for LightweightMR reconstruction.
Usage:
python -m lightweightmr --input path/to/pointcloud.ply --output path/to/mesh.ply
"""
import argparse
import os
import sys
from .optimize import Runner
def main():
parser = argparse.ArgumentParser(description="LightweightMR — pure-Python mesh reconstruction")
parser.add_argument("--input", "-i", required=True, help="Input point cloud (.ply, .pcd, .xyz)")
parser.add_argument("--output", "-o", required=True, help="Output mesh file (.ply or .obj)")
parser.add_argument("--device", default="cpu", help="torch device: cpu or cuda")
parser.add_argument("--out-dir", default="./output", help="Intermediate outputs directory")
parser.add_argument("--sdf-iters", type=int, default=20_000, help="SDF training iterations")
parser.add_argument("--vg-iters", type=int, default=8_000, help="Vertex generation iterations")
parser.add_argument("--sdf-lr", type=float, default=1e-3, help="SDF learning rate")
parser.add_argument("--vg-lr", type=float, default=1e-3, help="VG learning rate")
parser.add_argument("--sdf-batch", type=int, default=5_000, help="SDF batch size")
parser.add_argument("--vertices", type=int, default=3_400, help="Target number of vertices")
parser.add_argument("--update-size", type=int, default=5, help="Curriculum update steps")
parser.add_argument("--update-ratio", type=float, default=1.2, help="Curriculum growth ratio")
parser.add_argument("--k-samples", type=int, default=21, help="Tetrahedron interior samples")
parser.add_argument("--multires", type=int, default=8, help="Positional encoding frequencies")
parser.add_argument("--queries-size", type=int, default=1_000_000, help="SDF query sample count")
parser.add_argument("--surface-queries", type=int, default=200_000, help="Surface query count")
parser.add_argument("--project-sdf-level", type=float, default=0.0, help="SDD projection level")
parser.add_argument("--save-freq", type=int, default=2_000, help="Checkpoint save frequency")
parser.add_argument("--resume-sdf", default=None, help="Resume from SDF checkpoint .pth")
args = parser.parse_args()
if not os.path.exists(args.input):
print(f"ERROR: input file not found: {args.input}")
sys.exit(1)
runner = Runner(
pointcloud_path=args.input,
out_dir=args.out_dir,
device=args.device,
sdf_iters=args.sdf_iters,
vg_iters=args.vg_iters,
sdf_lr=args.sdf_lr,
vg_lr=args.vg_lr,
sdf_batch=args.sdf_batch,
vertices_size=args.vertices,
update_size=args.update_size,
update_ratio=args.update_ratio,
k_samples=args.k_samples,
multires=args.multires,
queries_size=args.queries_size,
surface_queries=args.surface_queries,
project_sdf_level=args.project_sdf_level,
save_freq=args.save_freq,
)
if args.resume_sdf:
print(f"Loading SDF checkpoint: {args.resume_sdf}")
runner.load_sdf_checkpoint(args.resume_sdf)
v, f = runner.run(mesh_path=args.output)
print(f"\nDone! Mesh: {len(v)} vertices, {len(f)} faces")
if __name__ == "__main__":
main()
|