Skip to content

Sync datasets with tristan-deep/OpenH-RF - #39

Merged
waltsims merged 164 commits into
open-h:mainfrom
tristan-deep:datasets-sync-2026-09-25
Sep 30, 2026
Merged

waltsims merged 164 commits into
open-h:mainfrom
tristan-deep:datasets-sync-2026-09-25

Conversation

@tristan-deep

@tristan-deep tristan-deep commented Sep 25, 2026 •

Copy link
Copy Markdown
Contributor

Syncs datasets/ with tristan-deep/OpenH-RF main: all data cards moved to the common layout, plus pipeline, script and figure updates across the 33 datasets. waterloo-largeartery is renamed to waterloo-femoralvein.

Only datasets/ changes (316 files). In sync with Hugging Face files.


Update — four more datasets. One commit per dataset brings these folders in line with tristan-deep/OpenH-RF and the Hub:

Dataset tristan-deep PR Hub PR
concordia #46 #77
weillcornell #33, #47 #62
stanford-murine #48 #78
resolvestroke #49 #79

Every Hub-hosted file in these folders is identical to the linked tristan-deep and Hub PRs; the scripts are GitHub-only. Pending: tristan-deep #48, #49 and Hub #62, #77, #78, #79 are not merged yet; once they are, all three are in sync. Still only datasets/ changes; the zea bump is #40.

Files changed: 22.

~ datasets/concordia/README.md
~ datasets/concordia/assets/classes.png
~ datasets/concordia/assets/reference_capture.png
~ datasets/concordia/reconstruct.py
~ datasets/weillcornell/README.md
~ datasets/weillcornell/pipeline.yaml
~ datasets/weillcornell/reconstruct.py
+ datasets/weillcornell/assets/ac10_15m_SK_pipeline.png
+ datasets/weillcornell/assets/hero.png
+ datasets/weillcornell/assets/main.png
- datasets/weillcornell/assets/ac1_15m_SK_pipeline.png
~ datasets/stanford-murine/README.md
~ datasets/stanford-murine/pipeline_hadamard.yaml
~ datasets/stanford-murine/pipeline_multifocal.yaml
~ datasets/stanford-murine/pipeline_synthetic_aperture.yaml
~ datasets/stanford-murine/reconstruct.py
+ datasets/stanford-murine/assets/hero.png
+ datasets/stanford-murine/assets/main.png
+ datasets/resolvestroke/clinical/assets/main.png
+ datasets/resolvestroke/phantom_flow/assets/main.png
+ datasets/resolvestroke/phantom_mp/assets/main.png
+ datasets/resolvestroke/saddle/assets/main.png

tristan-deep and others added 30 commits May 18, 2026 17:34
# Conflicts:
#	examples/templates/verasonics/convert.py
#	examples/templates/verasonics/pipeline.yaml
#	examples/templates/verasonics/reconstruct.py
#	uv.lock
One directory per sub-dataset of nvidia/OpenH-RF on the Hugging Face Hub,
each with the data card as published, the pipeline YAML, and a runnable
reconstruction script whose inputs stream from the Hub.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Reconstruction scripts write PNGs and GIFs next to themselves, so ignore *.gif
alongside the *.png rule that was already there.

A data card's artwork is the exception: it is checked in, under an assets/
folder beside the card, as datasets/tue-carotid/assets does. The negation is
depth-agnostic because submissions nest, e.g. datasets/ubc/module_A.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Updated contributor information and contact details for the dataset.
tristan-deep and others added 24 commits September 24, 2026 13:47
- reconstruct.py: house layout (flat main, ZEA_FILE/CONFIG/FRAME/OUT/HF_CONFIG,
  local per-track CONFIG, file closed before beamforming, default device);
  output unchanged (pixel-identical on CPU)
- utils.py: removed; only reconstruct.py used it
- pipeline/*.yaml: add parameters block (t_peak from custom/track_i_t_peak,
  xlims, dynamic_range) so zea process reproduces the B-mode exactly
- README.md: add zea process block, update processing section
- reconstruct.py: define the constants it uses (ZEA_FILE, N_FRAMES, OUT were
  undefined after the rename); standard names, OUT in assets/, HF_CONFIG,
  write_config(), dynamic range moved into PARAMETERS, drop DEVICE
- pipeline.yaml: regenerated, adds dynamic_range [-20, 0]
- assets/: add the 1000 kPa example output, drop the unreferenced 10 kPa one
- README.md: restore title, fix hero/schematic paths, add captions, correct
  the Data Validation paragraph (output path, no --output flag)
The Hub folder moves from waterloo-largeartery/ to waterloo-femoralvein/ in
HF PR #47; the card's sample link, zea process command and the script's
ZEA_FILE / HF_CONFIG / dataset link follow.
waterloo-femoralvein: point Hub paths at waterloo-femoralvein
Updated the dataset README to correct the raw RF channel count for the Twente vortexflow acquisition from 192 to 1. This aligns the documentation with the actual acquisition metadata and prevents confusion when processing or interpreting the dataset.
… YAML

- reconstruct.py: drop the input-path parsing and the unused build_pipeline()
  fallback; read everything inside the File context, beamform and plot after
  it closes; standard constants; output pixel-identical (zea 0.1.6 and main)
- pipeline_*.yaml: grid_size 1536 x 480 (0.104 mm isotropic, what the script
  rendered; was 1600 x 700, anisotropic) and dynamic_range [-50, 0]
build_pipeline() defines the delay-and-sum chain and write_config() writes it
with PARAMETERS (grid, dynamic range) to both track YAMLs, which the run then
loads. Output unchanged (pixel-identical).
weizmann-sampl: list individual contributors
ulmshare: tidy reconstruct.py, add hero image
Twente-VortexFlow: Corrections for IUS 2026 submission
unc-openpros: add license conversion agreement in README
ulmshare: stream raw data in large blocks instead of one request per frame
@tristan-deep

Copy link
Copy Markdown
Contributor Author

Follow up of #33

@waltsims
waltsims merged commit 90b18de into open-h:main Sep 30, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.