Published March 12, 2026 | Version v1
Dataset Open

EmbryoTempoFormer Princeton External-Domain Processed Bundle (sbiad840_aligned_4p5)

Authors/Creators

Contributors

Contact person:

Description

This archive provides the optional processed Princeton external-domain bundle used in the EmbryoTempoFormer (ETF) revision for cross-site evaluation and low-shot site-specific adaptation.

Archive name
- sbiad840_aligned_4p5.tar.gz

Purpose
This is a companion archive to the main ETF reproducibility bundle hosted at:
https://zenodo.org/records/18318139

Unlike the main bundle, which is sufficient to reproduce the primary paper results based on S-BIAD531, this companion archive is only required for reproducing the Princeton external-domain analyses (cross-site zero-shot transfer, dense single-frame comparison, and low-shot adaptation).

Contents
After extraction, the archive provides:

sbiad840_aligned_4p5/
  processed_28C5_sbiad840/
  processed_25C_sbiad840/
  splits/

Details
- processed_28C5_sbiad840/: ETF-compatible uint8 NumPy arrays (*.npy), one per embryo, aligned to T0 = 4.5 hpf, for the Princeton 28.5°C condition
- processed_25C_sbiad840/: ETF-compatible uint8 NumPy arrays (*.npy), one per embryo, aligned to T0 = 4.5 hpf, for the Princeton 25°C condition
- splits/: JSON split files for Princeton external-domain evaluation and fine-tuning workflows

Preprocessing notes
These arrays were generated from the public Princeton PNG time-folder export from BioImage Archive accession S-BIAD840. The preprocessing aligns both conditions to the ETF nominal time base (T0 = 4.5 hpf, DT = 0.25 h), applies percentile intensity clipping, resizes frames to 384 × 384, and preserves the native aligned sequence length (168 frames) by default rather than padding to 192.

Expected extraction location
For the repository scripts to work with default paths, extract this archive inside the ETF repository as:

<repo>/data/sbiad840_aligned_4p5/

so that the layout becomes:

data/sbiad840_aligned_4p5/
  processed_28C5_sbiad840/
  processed_25C_sbiad840/
  splits/

Recommended usage
After extracting this archive into the repository, the Princeton workflows can be run using the documented scripts in the codebase, including:
- scripts/35_aggregate_sbiad840.sh
- scripts/36_summarize_sbiad840.sh
- scripts/38_compare_sbiad840_kimmelnet.sh
- scripts/42_summarize_sbiad840_finetune.sh
- scripts/43_eval_sbiad840_finetuned.sh

Relationship to the main ETF bundle
Main ETF reproducibility bundle:
https://zenodo.org/records/18318139

This companion archive supplements that main bundle by providing the processed external-domain Princeton data required for the revision’s cross-site analyses.

Files

Files (3.6 GB)

Name Size Download all
md5:e2966c8afad0cc6b3b25f42639508e25
3.6 GB Download