The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
file: string
sha256: string
bytes: int64
iteration: int64
stage: string
smoke_only: bool
optimizer_included: bool
target_eval_parent_tiles: int64
target_train_images: int64
files: list<item: struct<path: string, sha256: string, bytes: int64>>
child 0, item: struct<path: string, sha256: string, bytes: int64>
child 0, path: string
child 1, sha256: string
child 2, bytes: int64
credentials_included: bool
format: int64
target_train_labels: bool
target_eval_images: int64
main_training_results_included: bool
purpose: string
source_training_images_included: bool
to
{'format': Value('int64'), 'purpose': Value('string'), 'files': List({'path': Value('string'), 'sha256': Value('string'), 'bytes': Value('int64')}), 'target_train_images': Value('int64'), 'target_eval_images': Value('int64'), 'target_eval_parent_tiles': Value('int64'), 'target_train_labels': Value('bool'), 'source_training_images_included': Value('bool'), 'main_training_results_included': Value('bool'), 'credentials_included': 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 2951, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, 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 547, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, 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
file: string
sha256: string
bytes: int64
iteration: int64
stage: string
smoke_only: bool
optimizer_included: bool
target_eval_parent_tiles: int64
target_train_images: int64
files: list<item: struct<path: string, sha256: string, bytes: int64>>
child 0, item: struct<path: string, sha256: string, bytes: int64>
child 0, path: string
child 1, sha256: string
child 2, bytes: int64
credentials_included: bool
format: int64
target_train_labels: bool
target_eval_images: int64
main_training_results_included: bool
purpose: string
source_training_images_included: bool
to
{'format': Value('int64'), 'purpose': Value('string'), 'files': List({'path': Value('string'), 'sha256': Value('string'), 'bytes': Value('int64')}), 'target_train_images': Value('int64'), 'target_eval_images': Value('int64'), 'target_eval_parent_tiles': Value('int64'), 'target_train_labels': Value('bool'), 'source_training_images_included': Value('bool'), 'main_training_results_included': Value('bool'), 'credentials_included': 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.
GSD-SFUDA development bundle
Self-contained code, prepared IRRG views, inference checkpoints and a Kaggle notebook for the first development stage of physical-scale-conditioned pseudo-label reliability. The primary experiment uses SND on the public SiamSeg-derived Vaihingen/Potsdam tile groups with a MiT-B2 implementation.
Download gsd_sfuda_kaggle.ipynb, upload it into Kaggle, enable Internet and select a T4 GPU. The notebook anonymously downloads this entire repository, pins its revision, verifies SHA-256, installs dependencies, runs geometry/loss tests, frozen-teacher diagnostics and optional two-step SND integration smoke. Only GPU0 is used. No Hugging Face token is required.
Included data and checkpoints
- 16 image-only target adaptation patches and 128 evaluation patches, distributed across the 14 Potsdam evaluation tiles.
- Native IRRG imagery and cached views at ratios1.2/1.25/1.8/2.0, created directly from original parent tiles. GSD9cm reference corresponds to1.8 against Potsdam5cm; generic2.0 is a strong fixed-scale control.
- Original eroded/noBoundary evaluation masks, class IDs0..5/ignore255.
- Source8,500-step inference snapshot and a2-step smoke source checkpoint. Optimizers/RNG states, raw download archives and source training images are excluded.
The subset is for research development, not final benchmark reporting. The8,500-step snapshot is incomplete relative to the40k source schedule. Source40k completion, warm20k and the four-arm1500-step pilot run on HPC; these later results are not represented by this initial package.
What is measured
Native labels are held fixed while confidence, generic disagreement and physical-scale disagreement rank pixels at matched predicted-class coverage. The script exports risk/coverage, conditional confidence-bin risks, joint correct/wrong counts, fusion predictions and reusable FP32 score caches. Offline lambda sweeps use CPU and need no additional GPU forwards.
Training and inference use continuous logits resampling followed by softmax, matching the native SND ordering. No hard-label interpolation is used for JS. Source/target geometry and valid support are stored in view_manifest.json. Adaptation never reads target train GT. Evaluation labels are read only in the declared diagnostic/evaluation process.
Reproducibility
bundle_manifest.json records every shipped file's size and SHA-256. The code archive contains its own per-file manifest. Runtime outputs record checkpoint hash, view-manifest hash, Torch version, device, wall time and VRAM. Kaggle smoke results must not be compared with full HPC training.
Attribution and original terms
Data originate from the ISPRS Vaihingen benchmark and ISPRS Potsdam benchmark. ISPRS acknowledges BSF Swissphoto for the Potsdam imagery. Data rights and original benchmark terms remain with the providers.
Protocol/code references: SiamSeg, SND and SegFormer. Upstream attribution is retained in the code; the SiamSeg Apache license is included. The complete NVIDIA SegFormer source license is included in references/segformer/LICENSE; its research/evaluation use restriction applies to the SegFormer-derived architecture. This bundle does not assign a new blanket license to the original data or upstream code.
- Downloads last month
- 48