Add multi-config dataset loading script
Browse files
NGS.py
ADDED
|
@@ -0,0 +1,95 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""NGS Transparency Datasets loading script"""
|
| 2 |
+
|
| 3 |
+
import datasets
|
| 4 |
+
from pathlib import Path
|
| 5 |
+
import os
|
| 6 |
+
|
| 7 |
+
_DESCRIPTION = "A collection of datasets (stones, objects, DTU, OmniObject3D) augmented with noise infills to benchmark 'false transparency' in 3DGS."
|
| 8 |
+
_HOMEPAGE = "https://huggingface.co/datasets/OpsiClear/NGS"
|
| 9 |
+
_LICENSE = "Mixed licenses. See dataset card for details."
|
| 10 |
+
_CITATION = """
|
| 11 |
+
@inproceedings{elhakie2025ngs,
|
| 12 |
+
title={Fix False Transparency by Noise Guided Splatting},
|
| 13 |
+
author={El Hakie, Aly and Lu, Yiren and Yin, Yu and Jenkins, Michael and Liu, Yehe},
|
| 14 |
+
booktitle={Conference on Neural Information Processing Systems (NeurIPS)},
|
| 15 |
+
year={2025}
|
| 16 |
+
}
|
| 17 |
+
"""
|
| 18 |
+
|
| 19 |
+
class NGSDatasetConfig(datasets.BuilderConfig):
|
| 20 |
+
"""BuilderConfig for the NGS Transparency Datasets."""
|
| 21 |
+
def __init__(self, **kwargs):
|
| 22 |
+
super().__init__(**kwargs)
|
| 23 |
+
|
| 24 |
+
class NGSDatasets(datasets.GeneratorBasedBuilder):
|
| 25 |
+
"""NGS Transparency Datasets Loader"""
|
| 26 |
+
|
| 27 |
+
VERSION = datasets.Version("1.0.0")
|
| 28 |
+
|
| 29 |
+
BUILDER_CONFIGS = [
|
| 30 |
+
NGSDatasetConfig(name="stones", description="High-resolution scans of stone specimens."),
|
| 31 |
+
NGSDatasetConfig(name="objects", description="Scans of various everyday objects."),
|
| 32 |
+
NGSDatasetConfig(name="dtu", description="Augmented subset of the DTU dataset."),
|
| 33 |
+
NGSDatasetConfig(name="omniobject3d", description="Augmented subset of the OmniObject3D dataset."),
|
| 34 |
+
]
|
| 35 |
+
|
| 36 |
+
DEFAULT_CONFIG_NAME = "stones"
|
| 37 |
+
|
| 38 |
+
def _info(self):
|
| 39 |
+
return datasets.DatasetInfo(
|
| 40 |
+
description=_DESCRIPTION,
|
| 41 |
+
features=datasets.Features({
|
| 42 |
+
"scan_id": datasets.Value("string"),
|
| 43 |
+
"subset": datasets.Value("string"),
|
| 44 |
+
"images": datasets.Sequence(datasets.Image()),
|
| 45 |
+
"masks": datasets.Sequence(datasets.Image()),
|
| 46 |
+
"infill_gaussians_path": datasets.Value("string"),
|
| 47 |
+
"surface_gaussians_path": datasets.Value("string"),
|
| 48 |
+
}),
|
| 49 |
+
homepage=_HOMEPAGE,
|
| 50 |
+
license=_LICENSE,
|
| 51 |
+
citation=_CITATION,
|
| 52 |
+
)
|
| 53 |
+
|
| 54 |
+
def _split_generators(self, dl_manager):
|
| 55 |
+
data_dir = dl_manager.download_and_extract(self.config.name)
|
| 56 |
+
return [
|
| 57 |
+
datasets.SplitGenerator(
|
| 58 |
+
name=datasets.Split.TRAIN,
|
| 59 |
+
gen_kwargs={"data_dir": Path(data_dir)},
|
| 60 |
+
),
|
| 61 |
+
]
|
| 62 |
+
|
| 63 |
+
def _generate_examples(self, data_dir):
|
| 64 |
+
subset_name = self.config.name
|
| 65 |
+
subset_path = data_dir / subset_name
|
| 66 |
+
|
| 67 |
+
scan_dirs = sorted([d for d in subset_path.iterdir() if d.is_dir()])
|
| 68 |
+
|
| 69 |
+
for idx, scan_dir in enumerate(scan_dirs):
|
| 70 |
+
images = []
|
| 71 |
+
masks = []
|
| 72 |
+
|
| 73 |
+
# Aggregate images from all camera sub-folders
|
| 74 |
+
image_base_path = scan_dir / "images"
|
| 75 |
+
if image_base_path.exists():
|
| 76 |
+
for camera_dir in sorted(image_base_path.iterdir()):
|
| 77 |
+
if camera_dir.is_dir():
|
| 78 |
+
images.extend(sorted([str(p) for p in camera_dir.glob("*.png")]))
|
| 79 |
+
images.extend(sorted([str(p) for p in camera_dir.glob("*.jpg")]))
|
| 80 |
+
|
| 81 |
+
# Aggregate masks from all camera sub-folders
|
| 82 |
+
mask_base_path = scan_dir / "masks"
|
| 83 |
+
if mask_base_path.exists():
|
| 84 |
+
for camera_dir in sorted(mask_base_path.iterdir()):
|
| 85 |
+
if camera_dir.is_dir():
|
| 86 |
+
masks.extend(sorted([str(p) for p in camera_dir.glob("*.png")]))
|
| 87 |
+
|
| 88 |
+
yield idx, {
|
| 89 |
+
"scan_id": scan_dir.name,
|
| 90 |
+
"subset": subset_name,
|
| 91 |
+
"images": images,
|
| 92 |
+
"masks": masks,
|
| 93 |
+
"infill_gaussians_path": str(scan_dir / "infill.ply"),
|
| 94 |
+
"surface_gaussians_path": str(scan_dir / "surface.ply"),
|
| 95 |
+
}
|