File size: 4,491 Bytes
ea66a99 | 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 | """Utilities for loading Gaussian models from compressed artifacts."""
from __future__ import annotations
import json
from pathlib import Path
from typing import Dict, Tuple
import numpy as np
import torch
def _load_latent_matrix(latents_path: Path) -> Tuple[np.ndarray, Dict[str, object]]:
from tools import rest_autoencoder as rest
with rest._load_latents_pack(latents_path) as latents_pack:
meta = json.loads(str(latents_pack["meta"]))
if meta.get("quant_bits"):
shape = tuple(meta.get("latents_shape") or (meta["count"], meta["latent_dim"]))
if meta.get("bitpack_format") == "bitplane_little":
latents_q = rest._bitunpack_latents(
latents_pack["latents_bitpack"],
shape,
meta["quant_bits"],
)
else:
latents_q = latents_pack["latents_q"]
latents = rest._dequantise_latents(
latents_q,
latents_pack["latents_min"],
latents_pack["latents_max"],
meta["quant_bits"],
)
else:
latents = latents_pack["latents"]
return latents.astype(np.float32), meta
def _apply_residual_to_arrays(arrays: Dict[str, np.ndarray], rest_names: Tuple[str, ...], codebook_path: Path, assignments_path: Path) -> None:
if not codebook_path.exists() or not assignments_path.exists():
return
codebook = np.load(codebook_path)
assignments = np.load(assignments_path)
residual = codebook[assignments]
for idx, name in enumerate(rest_names):
base = arrays.get(name)
if base is None:
base = np.zeros(residual.shape[0], dtype=np.float32)
arrays[name] = base + residual[:, idx]
def decode_compressed_iteration(
model_path: Path,
iteration: int,
*,
device: str = "cuda",
chunk: int = -1,
) -> Dict[str, np.ndarray]:
"""Return field arrays reconstructed from compressed assets."""
point_dir = model_path / "point_cloud" / f"iteration_{iteration}"
compressed_dir = point_dir / "compressed"
if not compressed_dir.exists():
raise FileNotFoundError(f"No compressed data at {compressed_dir}")
pack_dir_lossless = compressed_dir / "lossless"
pack_dir_restfree = compressed_dir / "restfree"
latents_path = compressed_dir / "latents.npz"
checkpoint_path = compressed_dir / "rest_autoencoder.pt"
if pack_dir_lossless.exists():
from utils.lossless_pack import unpack_ply_to_arrays
arrays, _ = unpack_ply_to_arrays(pack_dir_lossless)
elif pack_dir_restfree.exists():
from tools import restfree_quant_pack as restfree
arrays, _ = restfree.unpack_restfree_to_arrays(pack_dir_restfree)
else:
raise FileNotFoundError(f"No lossless or restfree pack found in {compressed_dir}")
latents, _ = _load_latent_matrix(latents_path)
from tools import rest_autoencoder as rest
model, target_mean, target_std, _, _, target_names, _ = rest._load_checkpoint(checkpoint_path, device)
count = latents.shape[0]
for name in target_names:
if name not in arrays:
arrays[name] = np.zeros(count, dtype=np.float32)
target_mean_t = torch.from_numpy(target_mean).to(device)
target_std_t = torch.from_numpy(target_std).to(device)
recon = np.empty((latents.shape[0], len(target_names)), dtype=np.float32)
chunk_size = latents.shape[0] if chunk <= 0 else chunk
with torch.no_grad():
for start in range(0, latents.shape[0], chunk_size):
end = min(start + chunk_size, latents.shape[0])
batch = torch.from_numpy(latents[start:end]).to(device)
rec = model.decode_batch(batch) * target_std_t + target_mean_t
recon[start:end] = rec.cpu().numpy()
for idx, name in enumerate(target_names):
arrays[name] = recon[:, idx]
_apply_residual_to_arrays(
arrays,
tuple(target_names),
compressed_dir / "residual_codebook.npy",
compressed_dir / "residual_assignments.npy",
)
return arrays
def load_gaussians_from_compressed(
gaussians,
model_path: Path,
iteration: int,
*,
device: str = "cuda",
chunk: int = -1,
) -> None:
"""Populate a GaussianModel from compressed iteration assets."""
arrays = decode_compressed_iteration(model_path, iteration, device=device, chunk=chunk)
gaussians.load_from_arrays(arrays)
|