fix(nifti): enable lazy loading for Nifti1ImageWrapper #2
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
parquet scan options and docs
* more parquet stream arg docs * minor * minor
less api calls when resolving data_files
set dev version
This commit will be squashed.
* fix polars cast_column issue * remove debug statements * cast large_strings to string for image handling
allow streaming hdf5 files
retry open hf file
* keep hffs cache in workers when streaming * bonus: reorder hffs args to improve caching
* Update document_dataset.mdx * Update document_dataset.mdx OCR
* Add custom suffix support to from_generator * Renamed a new arg to fingerprint * Changed name to config_id in builder * Change version * Added a test * Version update * Update version * Update tests/test_arrow_dataset.py * Rename config_id to fingerprint in generator.py * Apply suggestions from code review * Update src/datasets/io/generator.py * Apply suggestions from code review --------- Co-authored-by: Quentin Lhoest <[email protected]>
* Add nifti support * update docs * update nifti after testing locally and from remote hub * update setup.py to add nibabel and update docs * add nifti_dataset * fix nifti dataset documentation * add nibabel to test dependency * Add section for creating a medical imaging dataset --------- Co-authored-by: Quentin Lhoest <[email protected]>
* WIP: shuffle working, interleave_ds not yet * remove debug statements * add test * update test * use recursive overwriting of generator seeds * update test description * remove debugging strings * return instances of baseexiterable instead of modifying inplace * add test to make sure multiple iterations over data are deterministic
* fix ci compressionfs * again * style
* update signature for _batch_setitems * arguments passthrough
…ggingface#7833) Adressing issue 7832
…huggingface#7831) * Fix argument passing in stratified shuffle split NumPy 2.0 changed the behavior of the `copy=False` parameter to be stricter. When `train_test_split` converted Arrow arrays to NumPy format for stratification, it triggered this error for non-contiguous arrays. Using `np.asarray()` allows copying when necessary, which is the recommended migration path per NumPy 2.0 documentation. * make style --------- Co-authored-by: Quentin Lhoest <[email protected]>
* add 3.14 * update ci * go home tf * torchcodec * numba * fix ci * no lz4 in python 3.14 * fix tests * again * again * again
* WIP: add audio, tests failing * WIP: add mono argument, tests failing * change from mono to num_channels in documentation, audio tests passing * update docs and move test for audio * update audio * update docstring for audio * Apply suggestions from code review --------- Co-authored-by: Quentin Lhoest <[email protected]>
fsspec 2025.10.0
release: 4.4.0
better streaming retries
…gingface#7848) remove mode parameter in docstring of pdf and video feature
* WIP: allow uploading of nifti * remove debug statements and fix test * remove debug statements * remove debug statements
Change arxiv to hg papers
* fix some broken links * some more --------- Co-authored-by: Quentin Lhoest <[email protected]>
* WIP: nifti vis working, now improve * seems to work fine, tests not there yet * remove uncommented lines
* try latest papaya * try niivue * update repr_html for nifti to work better with niivue * remove papaya files * remove papaya from setup.py * use ipyniivue * update nifti feature to use ipyniivue * add 3d crosshair for orientation * remove docstring
- Change dataobj initialization from get_fdata() to .dataobj - Preserves nibabel's ArrayProxy for lazy loading - Prevents OOM errors on large 4D fMRI files - Defers I/O errors to usage time for better error handling Resolves #1
Author
|
Apologies for the accidental close - was cleaning up branches on my fork and it auto-closed linked PRs. This lazy loading fix should integrate cleanly with your NiiVue work that just merged upstream. |
Author
|
Closing - reviewed directly on upstream PR huggingface#7887. Thanks! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
12 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
Single-line fix:
dataobj=nifti_image.get_fdata()→dataobj=nifti_image.dataobjPreserves nibabel's
ArrayProxyfor true lazy loading instead of eagerly loading entire NIfTI files into memory.Why
Discovered while testing BIDS loader with OpenNeuro data - a corrupted file caused an
EOFErrorwith no indication of which file was bad. The eagerget_fdata()call was loading data at decode time instead of access time.Also opened on upstream
huggingface#7887