Create RefRef.py
Browse files
RefRef.py
ADDED
@@ -0,0 +1,94 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import os
|
3 |
+
import datasets
|
4 |
+
|
5 |
+
_CITATION = """\
|
6 |
+
@InProceedings{...},
|
7 |
+
title = {Your Dataset Title},
|
8 |
+
author={Your Name},
|
9 |
+
year={2025}
|
10 |
+
}
|
11 |
+
"""
|
12 |
+
|
13 |
+
_DESCRIPTION = """\
|
14 |
+
Dataset containing multi-view images with camera poses, depth maps, and masks for NeRF training.
|
15 |
+
"""
|
16 |
+
|
17 |
+
_LICENSE = "MIT"
|
18 |
+
|
19 |
+
class RefRef(datasets.GeneratorBasedBuilder):
|
20 |
+
"""A dataset loader for NeRF-style data with camera poses, depth maps, and masks."""
|
21 |
+
|
22 |
+
VERSION = datasets.Version("1.0.0")
|
23 |
+
|
24 |
+
BUILDER_CONFIGS = [
|
25 |
+
datasets.BuilderConfig(
|
26 |
+
name="default",
|
27 |
+
version=VERSION,
|
28 |
+
description="Default configuration for NeRF dataset"
|
29 |
+
),
|
30 |
+
datasets.BuilderConfig(
|
31 |
+
name="ball",
|
32 |
+
version=VERSION,
|
33 |
+
description="Default configuration for NeRF dataset"
|
34 |
+
),
|
35 |
+
datasets.BuilderConfig(
|
36 |
+
name="ampoule",
|
37 |
+
version=VERSION,
|
38 |
+
description="Default configuration for NeRF dataset"
|
39 |
+
)
|
40 |
+
]
|
41 |
+
|
42 |
+
def _info(self):
|
43 |
+
features = datasets.Features({
|
44 |
+
"image": datasets.Image(),
|
45 |
+
"depth": datasets.Image(),
|
46 |
+
"mask": datasets.Image(),
|
47 |
+
"transform_matrix": datasets.Sequence(
|
48 |
+
datasets.Sequence(datasets.Value("float64"), length=4),
|
49 |
+
length=4
|
50 |
+
),
|
51 |
+
"rotation": datasets.Value("float32")
|
52 |
+
})
|
53 |
+
|
54 |
+
return datasets.DatasetInfo(
|
55 |
+
description=_DESCRIPTION,
|
56 |
+
features=features,
|
57 |
+
homepage="",
|
58 |
+
license=_LICENSE,
|
59 |
+
citation=_CITATION
|
60 |
+
)
|
61 |
+
|
62 |
+
def _split_generators(self, dl_manager):
|
63 |
+
# Automatically find all JSON files matching the split patterns
|
64 |
+
return [
|
65 |
+
datasets.SplitGenerator(
|
66 |
+
name=scene,
|
67 |
+
gen_kwargs={
|
68 |
+
"filepaths": os.path.join(f"https://huggingface.co/datasets/yinyue27/RefRef/resolve/main/image_data/textured_cube_scene/single-convex/{scene}/"),
|
69 |
+
"split": split
|
70 |
+
},
|
71 |
+
) for scene in ["ball", "ball_coloured", "cube", "cube_coloured"]
|
72 |
+
]
|
73 |
+
|
74 |
+
def _generate_examples(self, filepaths, split):
|
75 |
+
for split in ["train", "val", "test"]:
|
76 |
+
filepaths = os.path.join(filepaths, f"transforms_{split}.json")
|
77 |
+
with open(filepaths, "r", encoding="utf-8") as f:
|
78 |
+
try:
|
79 |
+
data = json.load(f)
|
80 |
+
except json.JSONDecodeError:
|
81 |
+
print("error")
|
82 |
+
|
83 |
+
scene_name = os.path.basename(os.path.dirname(filepaths))
|
84 |
+
|
85 |
+
for frame_idx, frame in enumerate(data.get("frames", [])):
|
86 |
+
base_dir = os.path.dirname(filepaths)
|
87 |
+
|
88 |
+
yield f"{scene_name}_{frame_idx}", {
|
89 |
+
"image": os.path.join(base_dir, frame["file_path"]+".png"),
|
90 |
+
"depth": os.path.join(base_dir, frame["depth_file_path"]),
|
91 |
+
"mask": os.path.join(base_dir, frame["mask_file_path"]),
|
92 |
+
"transform_matrix": frame["transform_matrix"],
|
93 |
+
"rotation": frame.get("rotation", 0.0)
|
94 |
+
}
|