Skip to content

Dtype resampling - #140

Open
robert-graf wants to merge 2 commits into
mainfrom
dtype_resampling
Open

Dtype resampling#140
robert-graf wants to merge 2 commits into
mainfrom
dtype_resampling

Conversation

@robert-graf

@robert-graf robert-graf commented Aug 26, 2026

Copy link
Copy Markdown
Collaborator

feat(nii): out_dtype for resampling + first-class float16/bool arrays

Summary

Adds an out_dtype kwarg to the NII resample surface so callers can pick the target dtype in one step, and teaches the NII wrapper to carry arrays in dtypes NIfTI-1 can't encode (float16, bool) without crashing — the upcast now happens only when we materialise a Nifti1Image (i.e. on save or when handing the underlying nibabel object out). Wires the new kwarg through the VibeSeg nnU-Net pre-processing pipeline so the input tensor fed to inference is 4× smaller.

_resample_from_to (TPTBox/core/internal/nii_help.py)

  • New out_dtype: np.dtype | type | str | None = None kwarg.
  • Forwarded to scipy.ndimage.affine_transform as output=, so the cast is folded into the interpolation write — no extra full-volume copy.
  • scipy.ndimage.affine_transform only accepts u8/u16/i16/i32/f32/f64 for output=. Requesting anything else (notably float16) transparently falls back to resampling into float32 and casting to the requested dtype in a single extra pass — still much cheaper than staying in the source dtype throughout.

NII wrapper wiring (TPTBox/core/nii_wrapper.py)

The out_dtype kwarg is threaded through every public entry point that flows into _resample_from_to:

┌──────────────────────────────────────────────┬──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ method │ note │
├──────────────────────────────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ rescale / rescale_ │ forwards to resample_from_to │
├──────────────────────────────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ resample_from_to / resample_from_to │ only affects the actual resample path; the identity-skip / reorient-only / pad-only shortcuts keep the source dtype (documented) │
├──────────────────────────────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ rescale_and_reorient / rescale_and_reorient_ │ forwards to rescale; the reorient step is a pure axis relabel and keeps the source dtype │
└──────────────────────────────────────────────┴──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┘

First-class float16 / bool inside a NII

Two new helpers colocated with _resample_from_to:

  • _NIFTI_UNSUPPORTED_DTYPE_UPCAST = {float16 → float32, bool → uint8}
  • _nifti_safe_dtype(dtype) — returns a nibabel-compatible dtype (identity for supported ones).
  • _arr_for_nifti1(arr) — returns arr as-is if the dtype is safe, else a copy in the upcast dtypeage is actually built.

Six call sites in nii_wrapper.py that used to raise HeaderDataError on float16/bool are guarded:

  • nii setter tuple-branch, nii setter no-header branch, nii getter reconstruction path (both Nifti1Image calls + set_data_dtype)
  • set_array — dropped the eager bool → uint8 / float16 → float32 upcasts; header uses _nifti_safe_dtype
  • set_dtype (both unpacked / non-unpacked branches) — header uses safe dtype; array keeps the re
  • save — on-disk copy goes through _arr_for_nifti1 and _nifti_safe_dtype; self._arr is untouched
  • _check_if_nifty_is_lying_about_its_dtype — deduplicated: the hardcoded float16 → float32 rule dtype, so future additions to the map propagate automatically

The caller-visible self.dtype reflects the actual array dtype (float16/bool) — upcasting is a ma

VibeSeg pre-inference (TPTBox/segmentation/VibeSeg/inference_nnunet.py)

  • Reordered squash_so_it_fits_in_float16 before the rescale (any needed intensity divide happens
  • Threaded the new kwarg: i.rescale_(zoom, mode=mode, verbose=True, out_dtype=np.float16).
  • If no zoom is provided, fall back to i.set_dtype(np.float16) so the pipeline always hands nnU-

Verified behaviour

rescale_(zoom, out_dtype=np.float16) → NII.dtype == float16, array bytes ÷4
save+reload float16 → file dtype = float32, max_abs_diff = 0.0
mask = (nii > 100) → NII.dtype == bool (in-memory)
save+reload bool → file dtype = uint8, values preserved
set_dtype(np.float16) → NII.dtype == float16

VibeSeg end-to-end (tutorial MR, RTX-class GPU, 3-fold ensemble)

Intermediate tensor handed to run_inference (measured with a spy on TPTBox.segmentation.nnUnet_utils.inference_api.run_inference):

┌───────────────────────────────────┬─────────┬────────────────────────────────┐
│ variant │ dtype │ array bytes │
├───────────────────────────────────┼─────────┼────────────────────────────────┤
│ baseline (source dtype preserved) │ float64 │ 40.29 MiB │
├───────────────────────────────────┼─────────┼────────────────────────────────┤
│ new (out_dtype=float16) │ float16 │ 10.07 MiB │
├───────────────────────────────────┼─────────┼────────────────────────────────┤
│ saved │ │ 30.22 MiB, −75% (4× reduction) │
└───────────────────────────────────┴─────────┴────────────────────────────────┘

Wall time: 7.8 s vs 7.9 s (≈ ±2%, GPU-inference dominates and hides the pre-processing delta). S: 99.9945% voxel-identical, no label with Dice < 0.999 — expected, because run_inference already
does an unconditional .astype(np.float16) on the input at inference_api.py:333, so the numbers rical between the two paths. The change simply earlier the cast and skips the float64 intermediates
in between (crop, pad, get_array).

The real payoff is on large clinical volumes: the same 4× saving is ~600 MiB on a 500×400×400 float64 input, which matters for the est_full_mb > total_ram_mb * 0.5 gate in run_inference_on_file that decides whether to fall
back to CPU-chunked inference.

🤖 Generated with Claude Code

@github-actions

github-actions Bot commented Aug 26, 2026

Copy link
Copy Markdown

Benchmark comparison

Speed (wall time per call)

baseline afc748d vs head afc748d · python 3.11.16 · 5 repeats + 1 warmup

Showing the 5 most-changed measurements per case; the rest are collapsed. Rows with a baseline below 3 ms are tagged (noise) — runner jitter on those swamps any real change.

ct_3d — shape (73, 47, 73)

Measurement baseline ms (median ±½·range) head ms (median ±½·range) Δ % p
nii_erode_msk 6.50 ±0.12 6.32 ±0.08 -2.8% 0.020
nii_load_img 10.98 ±0.10 10.78 ±0.07 -1.9% 0.025
nii_dilate_msk 126.06 ±0.56 123.83 ±0.45 -1.8% 0.000
nii_filter_connected_components 4.46 ±0.03 4.40 ±0.06 -1.5% 0.403
nii_fill_holes 4.72 ±0.06 4.66 ±0.12 -1.3% 0.543
… 31 more measurements
Measurement baseline ms (median ±½·range) head ms (median ±½·range) Δ % p
nii_rescale_seg 3.26 ±0.05 3.22 ±0.11 -1.1% 0.589
poi_calc_poi_from_subreg_vert 12.77 ±0.14 12.90 ±0.26 +1.1% 0.638
poi_calc_centroids_nocrop 4.39 ±0.10 4.36 ±0.07 -0.8% 0.407
nii_resample_from_to 42.04 ±0.09 41.99 ±0.32 -0.1% 0.817
nii_save 11.79 ±0.16 11.80 ±0.07 +0.1% 0.807
nii_rescale 39.89 ±0.22 39.90 ±0.31 +0.0% 0.501
metric_voxels 250463.00 ±0.00 250463.00 ±0.00 +0.0% —
metric_labels 3.00 ±0.00 3.00 ±0.00 +0.0% —
metric_foreground_pct 14.51 ±0.00 14.51 ±0.00 +0.0% —
nii_extract_label 0.49 ±0.02 0.42 ±0.03 -14.1% (noise) 0.002
nii_apply_crop 0.78 ±0.04 0.68 ±0.06 -12.5% (noise) 0.012
nii_get_array 0.06 ±0.01 0.05 ±0.00 -8.6% (noise) 0.048
poi_rescale 0.24 ±0.02 0.25 ±0.03 +6.7% (noise) 0.867
nii_compute_crop 0.35 ±0.01 0.33 ±0.02 -6.5% (noise) 0.082
nii_pad_to 0.78 ±0.04 0.73 ±0.06 -6.4% (noise) 0.073
nii_set_dtype 0.55 ±0.02 0.51 ±0.03 -6.0% (noise) 0.008
nii_unique 0.72 ±0.01 0.68 ±0.02 -5.1% (noise) 0.002
nii_map_labels 1.08 ±0.02 1.03 ±0.02 -4.7% (noise) 0.001
poi_load 1.00 ±0.04 1.04 ±0.08 +3.9% (noise) 0.289
poi_resample_from_to 0.74 ±0.03 0.71 ±0.04 -3.4% (noise) 0.732
poi_map_labels 0.19 ±0.02 0.20 ±0.02 +2.9% (noise) 0.928
poi_calc_centroids 2.57 ±0.07 2.51 ±0.05 -2.5% (noise) 0.522
poi_save 0.47 ±0.03 0.48 ±0.03 +2.2% (noise) 0.381
nii_get_connected_components 2.32 ±0.02 2.27 ±0.03 -2.1% (noise) 0.022
nii_center_of_masses 1.76 ±0.01 1.73 ±0.01 -1.8% (noise) 0.000
nii_volumes 1.91 ±0.03 1.88 ±0.02 -1.5% (noise) 0.055
nii_load_seg 2.88 ±0.09 2.83 ±0.12 -1.5% (noise) 0.204
poi_local_to_global_arr 0.32 ±0.02 0.33 ±0.01 +1.4% (noise) 0.991
nii_reorient 0.87 ±0.05 0.86 ±0.06 -0.9% (noise) 0.512
poi_reorient 0.43 ±0.03 0.44 ±0.03 +0.8% (noise) 0.727
poi_to_global 0.23 ±0.01 0.23 ±0.01 +0.5% (noise) 0.836

ct_2d — shape (73, 47, 1)

Measurement baseline ms (median ±½·range) head ms (median ±½·range) Δ % p
nii_resample_from_to 7.78 ±0.11 7.38 ±0.13 -5.1% 0.001
nii_rescale 5.78 ±0.14 5.50 ±0.18 -5.0% 0.016
nii_dilate_msk 14.94 ±0.09 14.49 ±0.06 -3.0% 0.000
metric_voxels 3431.00 ±0.00 3431.00 ±0.00 +0.0% —
metric_labels 3.00 ±0.00 3.00 ±0.00 +0.0% —
… 30 more measurements
Measurement baseline ms (median ±½·range) head ms (median ±½·range) Δ % p
metric_foreground_pct 37.60 ±0.00 37.60 ±0.00 +0.0% —
nii_apply_crop 0.76 ±0.03 0.69 ±0.02 -8.5% (noise) 0.002
poi_save 0.52 ±0.02 0.56 ±0.05 +8.5% (noise) 0.248
nii_unique 0.14 ±0.02 0.13 ±0.01 -8.4% (noise) 0.137
nii_get_array 0.02 ±0.00 0.02 ±0.00 -7.1% (noise) 0.207
nii_compute_crop 0.16 ±0.01 0.15 ±0.00 -3.9% (noise) 0.178
nii_map_labels 0.53 ±0.02 0.55 ±0.02 +3.8% (noise) 0.097
nii_get_connected_components 0.70 ±0.03 0.68 ±0.01 -3.7% (noise) 0.091
poi_calc_centroids_nocrop 1.24 ±0.05 1.20 ±0.02 -3.3% (noise) 0.038
nii_set_dtype 0.46 ±0.01 0.47 ±0.04 +3.2% (noise) 0.330
poi_to_global 0.26 ±0.01 0.25 ±0.01 -3.1% (noise) 0.053
nii_reorient 0.89 ±0.04 0.91 ±0.04 +2.9% (noise) 0.591
nii_extract_label 0.46 ±0.01 0.48 ±0.02 +2.6% (noise) 0.336
nii_filter_connected_components 0.88 ±0.03 0.91 ±0.05 +2.4% (noise) 0.195
nii_load_img 1.63 ±0.07 1.60 ±0.05 -2.4% (noise) 0.384
poi_reorient 0.49 ±0.02 0.50 ±0.01 +2.3% (noise) 0.243
nii_erode_msk 1.04 ±0.01 1.06 ±0.03 +2.2% (noise) 0.111
nii_rescale_seg 0.85 ±0.03 0.84 ±0.02 -2.1% (noise) 0.187
nii_load_seg 1.76 ±0.09 1.79 ±0.07 +1.9% (noise) 0.944
nii_volumes 0.31 ±0.01 0.30 ±0.01 -1.8% (noise) 0.242
nii_center_of_masses 0.22 ±0.01 0.22 ±0.01 -1.8% (noise) 0.200
nii_fill_holes 0.94 ±0.03 0.93 ±0.02 -1.0% (noise) 0.904
nii_pad_to 0.61 ±0.02 0.61 ±0.03 +1.0% (noise) 0.557
poi_calc_centroids 1.00 ±0.05 0.99 ±0.01 -1.0% (noise) 0.152
nii_save 1.47 ±0.03 1.46 ±0.08 -1.0% (noise) 0.944
poi_load 1.10 ±0.03 1.11 ±0.03 +0.9% (noise) 0.440
poi_rescale 0.28 ±0.00 0.28 ±0.02 +0.7% (noise) 0.392
poi_map_labels 0.22 ±0.01 0.22 ±0.01 -0.3% (noise) 0.715
poi_local_to_global_arr 0.34 ±0.02 0.34 ±0.01 -0.2% (noise) 0.418
poi_resample_from_to 0.79 ±0.01 0.79 ±0.02 -0.1% (noise) 0.803

mri_3d — shape (68, 52, 67)

Measurement baseline ms (median ±½·range) head ms (median ±½·range) Δ % p
nii_load_seg 3.07 ±0.10 2.96 ±0.06 -3.8% 0.024
nii_filter_connected_components 4.91 ±0.06 4.74 ±0.07 -3.5% 0.012
nii_fill_holes 5.64 ±0.06 5.48 ±0.06 -2.7% 0.002
nii_dilate_msk 164.09 ±1.35 159.94 ±1.04 -2.5% 0.001
poi_calc_centroids_nocrop 8.95 ±0.14 8.76 ±0.13 -2.1% 0.017
… 31 more measurements
Measurement baseline ms (median ±½·range) head ms (median ±½·range) Δ % p
nii_rescale 38.11 ±0.15 37.57 ±0.35 -1.4% 0.013
nii_resample_from_to 40.30 ±0.44 39.74 ±0.49 -1.4% 0.090
nii_rescale_seg 3.26 ±0.07 3.22 ±0.11 -1.3% 0.828
nii_erode_msk 7.99 ±0.10 7.90 ±0.13 -1.0% 0.228
poi_calc_poi_from_subreg_vert 12.28 ±0.04 12.18 ±0.19 -0.7% 0.165
nii_load_img 9.01 ±0.07 8.98 ±0.08 -0.3% 0.616
nii_save 9.65 ±0.07 9.64 ±0.17 -0.1% 0.466
metric_voxels 236912.00 ±0.00 236912.00 ±0.00 +0.0% —
metric_labels 8.00 ±0.00 8.00 ±0.00 +0.0% —
metric_foreground_pct 18.21 ±0.00 18.21 ±0.00 +0.0% —
nii_get_array 0.06 ±0.01 0.05 ±0.00 -10.0% (noise) 0.097
poi_local_to_global_arr 0.36 ±0.01 0.34 ±0.01 -7.5% (noise) 0.003
poi_save 0.58 ±0.02 0.62 ±0.02 +6.1% (noise) 0.039
nii_map_labels 1.09 ±0.04 1.03 ±0.02 -5.2% (noise) 0.005
poi_map_labels 0.27 ±0.03 0.26 ±0.01 -4.9% (noise) 0.248
poi_to_global 0.31 ±0.01 0.29 ±0.01 -4.4% (noise) 0.357
poi_rescale 0.32 ±0.01 0.33 ±0.02 +3.9% (noise) 0.181
poi_calc_centroids 2.89 ±0.07 2.79 ±0.03 -3.7% (noise) 0.022
nii_unique 0.70 ±0.02 0.67 ±0.02 -3.5% (noise) 0.248
nii_pad_to 0.79 ±0.02 0.77 ±0.03 -3.4% (noise) 0.069
poi_load 2.27 ±0.03 2.33 ±0.03 +2.5% (noise) 0.034
nii_apply_crop 0.79 ±0.04 0.77 ±0.01 -2.2% (noise) 0.253
nii_volumes 2.01 ±0.02 1.97 ±0.02 -1.8% (noise) 0.054
nii_extract_label 0.52 ±0.02 0.51 ±0.02 -1.7% (noise) 0.922
poi_reorient 0.52 ±0.01 0.53 ±0.06 +1.7% (noise) 0.366
poi_resample_from_to 0.86 ±0.03 0.84 ±0.07 -1.6% (noise) 0.874
nii_set_dtype 0.57 ±0.02 0.58 ±0.02 +1.2% (noise) 0.690
nii_compute_crop 0.36 ±0.02 0.36 ±0.01 -0.9% (noise) 0.265
nii_center_of_masses 2.01 ±0.02 1.99 ±0.04 -0.8% (noise) 0.846
nii_get_connected_components 2.28 ±0.03 2.26 ±0.07 -0.8% (noise) 0.642
nii_reorient 0.98 ±0.05 0.98 ±0.18 +0.0% (noise) 0.431

mri_2d — shape (68, 52, 1)

Measurement baseline ms (median ±½·range) head ms (median ±½·range) Δ % p
nii_resample_from_to 7.88 ±0.10 7.59 ±0.14 -3.6% 0.003
nii_dilate_msk 19.85 ±0.07 19.23 ±0.07 -3.1% 0.000
nii_rescale 5.49 ±0.10 5.42 ±0.08 -1.2% 0.150
metric_voxels 3536.00 ±0.00 3536.00 ±0.00 +0.0% —
metric_labels 7.00 ±0.00 7.00 ±0.00 +0.0% —
… 30 more measurements
Measurement baseline ms (median ±½·range) head ms (median ±½·range) Δ % p
metric_foreground_pct 50.17 ±0.00 50.17 ±0.00 +0.0% —
nii_volumes 0.49 ±0.03 0.45 ±0.02 -6.4% (noise) 0.079
nii_rescale_seg 0.89 ±0.14 0.83 ±0.02 -6.3% (noise) 0.141
nii_load_seg 1.81 ±0.03 1.91 ±0.09 +5.4% (noise) 0.432
poi_calc_centroids 1.18 ±0.01 1.24 ±0.08 +4.9% (noise) 0.152
nii_set_dtype 0.42 ±0.01 0.44 ±0.02 +4.7% (noise) 0.218
nii_reorient 0.93 ±0.02 0.89 ±0.04 -4.3% (noise) 0.483
poi_local_to_global_arr 0.35 ±0.01 0.34 ±0.02 -4.2% (noise) 0.054
nii_get_array 0.02 ±0.00 0.02 ±0.00 +4.1% (noise) 0.755
nii_pad_to 0.63 ±0.02 0.61 ±0.01 -4.0% (noise) 0.028
nii_fill_holes 1.19 ±0.04 1.15 ±0.02 -3.4% (noise) 0.091
nii_get_connected_components 0.68 ±0.02 0.66 ±0.03 -3.4% (noise) 0.195
nii_center_of_masses 0.41 ±0.02 0.42 ±0.02 +3.2% (noise) 0.884
nii_extract_label 0.47 ±0.05 0.46 ±0.01 -2.6% (noise) 0.212
poi_save 0.55 ±0.02 0.57 ±0.03 +2.5% (noise) 0.677
nii_filter_connected_components 0.95 ±0.02 0.93 ±0.01 -2.4% (noise) 0.313
poi_calc_centroids_nocrop 1.51 ±0.07 1.48 ±0.03 -2.1% (noise) 0.146
nii_load_img 1.58 ±0.06 1.55 ±0.12 -1.9% (noise) 0.570
poi_to_global 0.28 ±0.01 0.29 ±0.00 +1.7% (noise) 0.646
poi_map_labels 0.26 ±0.02 0.27 ±0.01 +1.6% (noise) 0.907
poi_load 2.00 ±0.03 1.97 ±0.04 -1.6% (noise) 0.204
poi_resample_from_to 0.81 ±0.03 0.80 ±0.03 -1.3% (noise) 0.285
nii_unique 0.14 ±0.00 0.13 ±0.00 -1.1% (noise) 0.210
poi_rescale 0.31 ±0.01 0.31 ±0.00 -1.1% (noise) 0.426
nii_apply_crop 0.76 ±0.02 0.76 ±0.04 -1.0% (noise) 0.671
nii_erode_msk 1.36 ±0.03 1.37 ±0.01 +1.0% (noise) 0.800
poi_reorient 0.52 ±0.02 0.51 ±0.02 -0.9% (noise) 0.669
nii_map_labels 0.53 ±0.03 0.53 ±0.04 +0.4% (noise) 0.864
nii_save 1.43 ±0.03 1.43 ±0.03 -0.4% (noise) 0.746
nii_compute_crop 0.16 ±0.01 0.16 ±0.01 -0.2% (noise) 0.334

Gate: a measurement fails when the baseline is ≥ 1 ms, the median grows by ≥ 50%, and Welch's t-test gives p < 0.05. metric_* rows are context only.

Memory (peak RSS growth per call)

baseline afc748d vs head afc748d · python 3.11.16 · 5 repeats + 1 warmup · sampler proc-status, isolation fork · measurement floor ≈ 0.90 MiB

Showing the 5 most-changed measurements per case; the rest are collapsed. Rows with a baseline below 3 MiB are tagged (noise) — runner jitter on those swamps any real change.

ct_3d — shape (73, 47, 73)

Measurement baseline MiB (median ±½·range) head MiB (median ±½·range) Δ % p
nii_extract_label 5.43 ±0.00 5.37 ±0.00 -1.2% —
nii_rescale_seg 6.01 ±0.00 5.95 ±0.00 -1.1% —
nii_rescale 6.24 ±0.00 6.17 ±0.00 -1.1% —
nii_resample_from_to 6.36 ±0.00 6.30 ±0.00 -1.0% —
nii_erode_msk 5.93 ±0.00 5.95 ±0.00 +0.3% —
… 31 more measurements
Measurement baseline MiB (median ±½·range) head MiB (median ±½·range) Δ % p
nii_load_img 3.57 ±0.00 3.56 ±0.00 -0.2% —
nii_load_seg 6.02 ±0.00 6.01 ±0.00 -0.1% —
nii_center_of_masses 3.28 ±0.00 3.27 ±0.00 -0.1% —
nii_volumes 3.53 ±0.00 3.52 ±0.00 -0.1% —
nii_save 3.71 ±0.00 3.71 ±0.00 -0.1% —
nii_set_dtype 5.25 ±0.00 5.24 ±0.00 -0.1% —
nii_pad_to 5.43 ±0.00 5.43 ±0.00 -0.1% —
nii_map_labels 5.50 ±0.00 5.49 ±0.00 -0.1% —
nii_reorient 5.91 ±0.00 5.91 ±0.00 -0.1% —
nii_dilate_msk 5.93 ±0.00 5.93 ±0.00 -0.1% —
nii_fill_holes 6.29 ±0.00 6.28 ±0.00 -0.1% —
nii_get_connected_components 6.62 ±0.00 6.62 ±0.00 -0.1% —
poi_calc_poi_from_subreg_vert 7.11 ±0.00 7.12 ±0.00 +0.1% —
nii_apply_crop 5.87 ±0.00 5.87 ±0.00 +0.0% —
nii_filter_connected_components 6.87 ±0.00 6.87 ±0.00 +0.0% —
poi_calc_centroids 6.06 ±0.00 6.06 ±0.00 +0.0% —
poi_calc_centroids_nocrop 5.93 ±0.00 5.93 ±0.00 +0.0% —
poi_reorient 3.84 ±0.00 3.84 ±0.00 +0.0% —
poi_to_global 3.09 ±0.00 3.09 ±0.00 +0.0% —
poi_resample_from_to 5.28 ±0.00 5.28 ±0.00 +0.0% —
metric_voxels 250463.00 ±0.00 250463.00 ±0.00 +0.0% —
metric_labels 3.00 ±0.00 3.00 ±0.00 +0.0% —
metric_foreground_pct 14.51 ±0.00 14.51 ±0.00 +0.0% —
nii_get_array 1.65 ±0.00 1.65 ±0.00 -0.2% (noise) —
nii_compute_crop 2.90 ±0.00 2.90 ±0.00 -0.1% (noise) —
nii_unique 2.96 ±0.00 2.96 ±0.00 -0.1% (noise) —
poi_rescale 2.61 ±0.00 2.61 ±0.00 +0.0% (noise) —
poi_local_to_global_arr 2.96 ±0.00 2.96 ±0.00 +0.0% (noise) —
poi_save 1.41 ±0.00 1.41 ±0.00 +0.0% (noise) —
poi_map_labels 1.78 ±0.00 1.78 ±0.00 +0.0% (noise) —
poi_load 1.62 ±0.00 1.62 ±0.00 +0.0% (noise) —

ct_2d — shape (73, 47, 1)

Measurement baseline MiB (median ±½·range) head MiB (median ±½·range) Δ % p
nii_extract_label 5.37 ±0.00 5.30 ±0.00 -1.2% —
nii_rescale_seg 5.95 ±0.00 5.88 ±0.00 -1.1% —
nii_rescale 6.17 ±0.00 6.11 ±0.00 -1.0% —
nii_resample_from_to 6.30 ±0.00 6.24 ±0.00 -0.9% —
nii_erode_msk 5.86 ±0.00 5.88 ±0.06 +0.3% 0.838
… 30 more measurements
Measurement baseline MiB (median ±½·range) head MiB (median ±½·range) Δ % p
nii_load_img 3.50 ±0.00 3.50 ±0.00 +0.0% —
nii_load_seg 5.95 ±0.00 5.95 ±0.00 +0.0% —
nii_save 3.65 ±0.00 3.65 ±0.00 +0.0% —
nii_set_dtype 5.18 ±0.00 5.18 ±0.00 +0.0% —
nii_reorient 5.84 ±0.00 5.84 ±0.00 +0.0% —
nii_apply_crop 5.80 ±0.00 5.80 ±0.00 +0.0% —
nii_pad_to 5.36 ±0.00 5.36 ±0.00 +0.0% —
nii_volumes 3.46 ±0.00 3.46 ±0.00 +0.0% —
nii_center_of_masses 3.21 ±0.00 3.21 ±0.00 +0.0% —
nii_map_labels 5.43 ±0.00 5.43 ±0.00 +0.0% —
nii_dilate_msk 5.87 ±0.00 5.87 ±0.00 +0.0% —
nii_fill_holes 6.22 ±0.00 6.22 ±0.00 +0.0% —
nii_get_connected_components 6.55 ±0.00 6.55 ±0.00 +0.0% —
nii_filter_connected_components 6.80 ±0.00 6.80 ±0.00 +0.0% —
poi_calc_centroids 5.99 ±0.00 5.99 ±0.00 +0.0% —
poi_calc_centroids_nocrop 5.93 ±0.00 5.93 ±0.00 +0.0% —
poi_reorient 3.77 ±0.00 3.77 ±0.00 +0.0% —
poi_to_global 3.02 ±0.00 3.02 ±0.00 +0.0% —
poi_resample_from_to 5.21 ±0.00 5.21 ±0.00 +0.0% —
metric_voxels 3431.00 ±0.00 3431.00 ±0.00 +0.0% —
metric_labels 3.00 ±0.00 3.00 ±0.00 +0.0% —
metric_foreground_pct 37.60 ±0.00 37.60 ±0.00 +0.0% —
nii_get_array 1.59 ±0.00 1.59 ±0.00 +0.0% (noise) —
nii_compute_crop 2.84 ±0.00 2.84 ±0.00 +0.0% (noise) —
nii_unique 2.90 ±0.00 2.90 ±0.00 +0.0% (noise) —
poi_rescale 2.54 ±0.00 2.54 ±0.00 +0.0% (noise) —
poi_local_to_global_arr 2.89 ±0.00 2.89 ±0.00 +0.0% (noise) —
poi_save 1.34 ±0.00 1.34 ±0.00 +0.0% (noise) —
poi_map_labels 1.71 ±0.00 1.71 ±0.00 +0.0% (noise) —
poi_load 1.55 ±0.00 1.55 ±0.00 +0.0% (noise) —

mri_3d — shape (68, 52, 67)

Measurement baseline MiB (median ±½·range) head MiB (median ±½·range) Δ % p
nii_extract_label 5.37 ±0.00 5.30 ±0.00 -1.2% —
nii_rescale_seg 5.95 ±0.00 5.88 ±0.00 -1.1% —
nii_rescale 6.17 ±0.00 6.11 ±0.00 -1.0% —
nii_resample_from_to 6.30 ±0.00 6.23 ±0.00 -1.0% —
nii_erode_msk 5.86 ±0.00 5.88 ±0.00 +0.3% —
… 31 more measurements
Measurement baseline MiB (median ±½·range) head MiB (median ±½·range) Δ % p
nii_load_img 3.56 ±0.00 3.56 ±0.00 +0.0% —
nii_load_seg 5.95 ±0.00 5.95 ±0.00 +0.0% —
nii_save 3.65 ±0.00 3.65 ±0.00 +0.0% —
nii_set_dtype 5.24 ±0.00 5.24 ±0.00 +0.0% —
nii_reorient 5.84 ±0.00 5.84 ±0.00 +0.0% —
nii_apply_crop 5.80 ±0.00 5.80 ±0.00 +0.0% —
nii_pad_to 5.36 ±0.00 5.36 ±0.00 +0.0% —
nii_volumes 3.46 ±0.00 3.46 ±0.00 +0.0% —
nii_center_of_masses 3.21 ±0.00 3.21 ±0.00 +0.0% —
nii_map_labels 5.43 ±0.00 5.43 ±0.00 +0.0% —
nii_dilate_msk 5.87 ±0.00 5.87 ±0.00 +0.0% —
nii_fill_holes 6.22 ±0.00 6.22 ±0.00 +0.0% —
nii_get_connected_components 6.55 ±0.00 6.55 ±0.00 +0.0% —
nii_filter_connected_components 6.80 ±0.00 6.80 ±0.00 +0.0% —
poi_calc_centroids 5.99 ±0.00 5.99 ±0.00 +0.0% —
poi_calc_centroids_nocrop 5.86 ±0.00 5.86 ±0.00 +0.0% —
poi_reorient 3.77 ±0.00 3.77 ±0.00 +0.0% —
poi_to_global 3.02 ±0.00 3.02 ±0.00 +0.0% —
poi_resample_from_to 5.21 ±0.00 5.21 ±0.00 +0.0% —
poi_calc_poi_from_subreg_vert 7.11 ±0.00 7.11 ±0.00 +0.0% —
metric_voxels 236912.00 ±0.00 236912.00 ±0.00 +0.0% —
metric_labels 8.00 ±0.00 8.00 ±0.00 +0.0% —
metric_foreground_pct 18.21 ±0.00 18.21 ±0.00 +0.0% —
nii_get_array 1.59 ±0.00 1.59 ±0.00 +0.0% (noise) —
nii_compute_crop 2.84 ±0.00 2.84 ±0.00 +0.0% (noise) —
nii_unique 2.90 ±0.00 2.90 ±0.00 +0.0% (noise) —
poi_rescale 2.54 ±0.00 2.54 ±0.00 +0.0% (noise) —
poi_local_to_global_arr 2.89 ±0.00 2.89 ±0.00 +0.0% (noise) —
poi_save 1.40 ±0.00 1.40 ±0.00 +0.0% (noise) —
poi_map_labels 1.71 ±0.00 1.71 ±0.00 +0.0% (noise) —
poi_load 1.55 ±0.00 1.55 ±0.00 +0.0% (noise) —

mri_2d — shape (68, 52, 1)

Measurement baseline MiB (median ±½·range) head MiB (median ±½·range) Δ % p
nii_extract_label 5.37 ±0.00 5.30 ±0.00 -1.2% —
nii_rescale_seg 5.95 ±0.00 5.88 ±0.00 -1.1% —
nii_rescale 6.17 ±0.00 6.11 ±0.00 -1.0% —
nii_resample_from_to 6.30 ±0.00 6.23 ±0.00 -1.0% —
nii_erode_msk 5.86 ±0.00 5.88 ±0.00 +0.3% —
… 30 more measurements
Measurement baseline MiB (median ±½·range) head MiB (median ±½·range) Δ % p
nii_load_img 3.56 ±0.00 3.56 ±0.00 +0.0% —
nii_load_seg 5.95 ±0.00 5.95 ±0.00 +0.0% —
nii_save 3.65 ±0.00 3.65 ±0.00 +0.0% —
nii_set_dtype 5.24 ±0.00 5.24 ±0.00 +0.0% —
nii_reorient 5.84 ±0.00 5.84 ±0.00 +0.0% —
nii_apply_crop 5.80 ±0.00 5.80 ±0.00 +0.0% —
nii_pad_to 5.36 ±0.00 5.36 ±0.00 +0.0% —
nii_volumes 3.46 ±0.00 3.46 ±0.00 +0.0% —
nii_center_of_masses 3.21 ±0.00 3.21 ±0.00 +0.0% —
nii_map_labels 5.43 ±0.00 5.43 ±0.00 +0.0% —
nii_dilate_msk 5.87 ±0.00 5.87 ±0.00 +0.0% —
nii_fill_holes 6.22 ±0.00 6.22 ±0.00 +0.0% —
nii_get_connected_components 6.55 ±0.00 6.55 ±0.00 +0.0% —
nii_filter_connected_components 6.80 ±0.00 6.80 ±0.00 +0.0% —
poi_calc_centroids 5.99 ±0.00 5.99 ±0.00 +0.0% —
poi_calc_centroids_nocrop 5.93 ±0.00 5.93 ±0.00 +0.0% —
poi_reorient 3.77 ±0.00 3.77 ±0.00 +0.0% —
poi_to_global 3.02 ±0.00 3.02 ±0.00 +0.0% —
poi_resample_from_to 5.21 ±0.00 5.21 ±0.00 +0.0% —
metric_voxels 3536.00 ±0.00 3536.00 ±0.00 +0.0% —
metric_labels 7.00 ±0.00 7.00 ±0.00 +0.0% —
metric_foreground_pct 50.17 ±0.00 50.17 ±0.00 +0.0% —
nii_get_array 1.59 ±0.00 1.59 ±0.00 +0.0% (noise) —
nii_compute_crop 2.84 ±0.00 2.84 ±0.00 +0.0% (noise) —
nii_unique 2.90 ±0.00 2.90 ±0.00 +0.0% (noise) —
poi_rescale 2.54 ±0.00 2.54 ±0.00 +0.0% (noise) —
poi_local_to_global_arr 2.89 ±0.00 2.89 ±0.00 +0.0% (noise) —
poi_save 1.40 ±0.00 1.40 ±0.00 +0.0% (noise) —
poi_map_labels 1.71 ±0.00 1.71 ±0.00 +0.0% (noise) —
poi_load 1.55 ±0.00 1.55 ±0.00 +0.0% (noise) —

Gate: a measurement fails when the baseline is ≥ 1 MiB, the median grows by ≥ 50%, and Welch's t-test gives p < 0.05. metric_* rows are context only.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@Hendrik-code Hendrik-code Awaiting requested review from Hendrik-code

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@robert-graf