Physically Grounded Monocular Depth via Nanophotonic Wavefront Encoding
Paper • 2503.15770 • Published
Datasets:
Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
version: int64
counts: struct<train: int64, test: int64>
child 0, train: int64
child 1, test: int64
depth_units: string
historical_test_also_used_for_validation: bool
source_manifest_sha256: struct<train: string, test: string>
child 0, train: string
child 1, test: string
samples: list<item: struct<split: string, id: string, files: list<item: struct<bytes: int64, sha256: string, (... 203 chars omitted)
child 0, item: struct<split: string, id: string, files: list<item: struct<bytes: int64, sha256: string, size: list< (... 191 chars omitted)
child 0, split: string
child 1, id: string
child 2, files: list<item: struct<bytes: int64, sha256: string, size: list<item: int64>, mode: string, format: strin (... 149 chars omitted)
child 0, item: struct<bytes: int64, sha256: string, size: list<item: int64>, mode: string, format: string, info_key (... 137 chars omitted)
child 0, bytes: int64
child 1, sha256: string
child 2, size: list<item: int64>
child 0, item: int64
child 3, mode: string
child 4, format: string
child 5, info_keys: list<item: null>
child 0, item: null
child 6, path: string
child 7, shape: list<item: int64>
child 0, item: int64
child 8, dtype: string
child 9, finite: bool
child 10, min: double
child 11, max: double
child 12, unique_values: int64
data_files: int64
bytes: int64
all_original_bytes_preserved: bool
sha256: string
filename: string
to
{'filename': Value('string'), 'bytes': Value('int64'), 'sha256': Value('string'), 'data_files': Value('int64'), 'counts': {'train': Value('int64'), 'test': Value('int64')}, 'all_original_bytes_preserved': Value('bool')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
version: int64
counts: struct<train: int64, test: int64>
child 0, train: int64
child 1, test: int64
depth_units: string
historical_test_also_used_for_validation: bool
source_manifest_sha256: struct<train: string, test: string>
child 0, train: string
child 1, test: string
samples: list<item: struct<split: string, id: string, files: list<item: struct<bytes: int64, sha256: string, (... 203 chars omitted)
child 0, item: struct<split: string, id: string, files: list<item: struct<bytes: int64, sha256: string, size: list< (... 191 chars omitted)
child 0, split: string
child 1, id: string
child 2, files: list<item: struct<bytes: int64, sha256: string, size: list<item: int64>, mode: string, format: strin (... 149 chars omitted)
child 0, item: struct<bytes: int64, sha256: string, size: list<item: int64>, mode: string, format: string, info_key (... 137 chars omitted)
child 0, bytes: int64
child 1, sha256: string
child 2, size: list<item: int64>
child 0, item: int64
child 3, mode: string
child 4, format: string
child 5, info_keys: list<item: null>
child 0, item: null
child 6, path: string
child 7, shape: list<item: int64>
child 0, item: int64
child 8, dtype: string
child 9, finite: bool
child 10, min: double
child 11, max: double
child 12, unique_values: int64
data_files: int64
bytes: int64
all_original_bytes_preserved: bool
sha256: string
filename: string
to
{'filename': Value('string'), 'bytes': Value('int64'), 'sha256': Value('string'), 'data_files': Value('int64'), 'counts': {'train': Value('int64'), 'test': Value('int64')}, 'all_original_bytes_preserved': Value('bool')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Metasurface Depth: real training and evaluation scenes
Real encoded image pairs and the corresponding original depth-label arrays used by the selected mixed-training checkpoints for Physically Grounded Monocular Depth via Nanophotonic Wavefront Encoding.
Contents
| Split | Scenes | Input PNGs | Depth NPYs | Stored resolution (height × width) |
|---|---|---|---|---|
train |
5 | 10 | 5 | 1200 × 1600 |
test |
42 | 84 | 42 | 1190 × 1596 |
The archive contains 141 original data files plus portable CSV manifests and a checksum/protocol manifest. Files are copied byte-for-byte: no resizing, re-normalization, re-simulation or label conversion is applied.
The split called test is the historical real evaluation list and was also
used as validation for selection of the released models. It is not an untouched
holdout test set. The five training scenes and 42 evaluation scenes have no
shared source paths or byte-identical ordered input pairs. This does not prove
absence of object/category overlap.
Format
Extract metasurface-real-v1.zip to a directory of your choice:
metasurface-real-v1/
train.csv
test.csv
manifest.json
samples/
train/<scene_id>/image1.png
train/<scene_id>/image2.png
train/<scene_id>/depth.npy
test/<scene_id>/...
CSV fields are id,image1,image2,depth, with paths relative to the CSV. The input
order must not be swapped. The model loader reads the PNGs as uint8 grayscale.
Depth files are finite float32 H×W arrays in meters, spatially aligned to both
images. Observed labels span approximately 0.25–1.10 m. Scene filenames may name
physical distances different from the stored label values; use the arrays as
provided rather than deriving labels from filenames.
These are processed inputs and supervision arrays used by the experiments, not a claim to publish all original sensor streams or independently measured dense RGB-D scans. The annotation/measurement procedure is not inferred from file names. The fourth field in the original experimental manifests duplicates image1; it is not a separate conventional RGB image and is not duplicated in this release.
Using the project code
After installing the associated metasurface_depth source release:
metasurface evaluate --manifest metasurface-real-v1/test.csv \
--weights checkpoints/small.pth --output runs/real-small.json
Point the real-data entry in your training configuration to train.csv, retaining
the selected synthetic/real mixture when reproducing that training setup. The
five real scenes are not the complete synthetic training set. They bypass
the simulator because the input pair is already captured/processed.
Training uses aligned random crops; evaluation uses aligned center crops to multiples of 14. The archived evaluation images already have dimensions divisible by 14. Ground truth is a training/evaluation label, never a neural input.
Provenance and limits
The release follows nano3d-train-5.txt and real_exp/nano3d-real.txt, the real
lists named by all three selected November checkpoints. Internal absolute paths
are excluded; file and source-list hashes are included for reproducibility.
The selected Base checkpoint does not reproduce every inspected paper-table
entry; the hypothesis that the table used synthetic-only training is unverified.
Scientific images from other source datasets, experiment logs, private paths and raw training checkpoints are not included.
License
Copyright 2026 the dataset contributors. Released under Creative Commons Attribution-NonCommercial 4.0 International. Noncommercial use, including academic research, sharing and adaptation is permitted with appropriate attribution, a link to the license, and an indication of changes. Commercial use is not permitted under this license. See LICENSE.md. This license covers the captured inputs, depth labels and dataset manifests; code and model weights retain their own licenses.
The authors authorized public noncommercial distribution on September 7, 2026. Please cite the paper when using this dataset.
- Downloads last month
- 11