OR-PAM-Reg-Temporal-26K: A Temporal Benchmark Dataset for OR-PAM Registration

Version

Current version: v2

Note: v1 of this dataset has been deprecated. Please use the current version (v2) for all experiments and benchmarks.

Overview

OR-PAM-Reg-Temporal-26K is a large-scale benchmark dataset for evaluating image registration methods in bidirectional optical-resolution photoacoustic microscopy (OR-PAM). Built from continuous frames of in vivo mouse brain vasculature, the dataset contains 26,550 paired image samples (512 × 512 pixels) and supports two registration tasks:

• Task 1 — Intra-frame registration: Align odd-line (forward scan) and even-line (backward scan) columns within each frame, correcting bidirectional scanning artifacts.
• Task 2 — Inter-frame temporal registration: Align the same spatial patch across different time frames for temporal motion correction.

Further details on data acquisition and processing will be provided in the accompanying methodology paper.

Dataset Statistics

Split	Samples	Percentage
Train	21,240	80%
Val	2,596	~10%
Test	2,714	~10%
Total	26,550	100%

Data Format

OR-PAM-Reg-Temporal-26K/
├── train.h5    # Training set (~10.5 GB)
├── val.h5      # Validation set (~1.3 GB)
├── test.h5     # Test set (~1.3 GB)
└── README.md   # This file

HDF5 Structure

Each .h5 file contains:

Key	Shape	Dtype	Description
odd	(N, 1, 512, 256)	float32	Odd-line (forward scan) images
even	(N, 1, 512, 256)	float32	Even-line (backward scan) images
frame_idx	(N,)	int32	Temporal frame index
patch_idx	(N,)	int32	Spatial patch position
sample_ids	(N,)	string	Unique identifiers (e.g., f001_p042)
metadata/	group	—	Processing parameters

• Pixel values: Normalized to [0, 1]
• Channel: Single-channel grayscale (photoacoustic signal intensity)
• Dimensions: (batch, channel, height, width) in PyTorch convention

Quick Start

import h5py
import torch
from torch.utils.data import Dataset

class ORPAMTemporalDataset(Dataset):
    def __init__(self, h5_path):
        with h5py.File(h5_path, 'r') as f:
            self.odd = f['odd'][:]
            self.even = f['even'][:]
            self.frame_idx = f['frame_idx'][:]
            self.patch_idx = f['patch_idx'][:]

    def __len__(self):
        return len(self.odd)

    def __getitem__(self, idx):
        odd = torch.from_numpy(self.odd[idx]).float()
        even = torch.from_numpy(self.even[idx]).float()
        return odd, even, self.frame_idx[idx], self.patch_idx[idx]

# Usage
dataset = ORPAMTemporalDataset('train.h5')
odd, even, fidx, pidx = dataset[0]
print(f'Odd: {odd.shape}, Even: {even.shape}, Frame: {fidx}, Patch: {pidx}')
# Output: Odd: torch.Size([1, 512, 256]), Even: torch.Size([1, 512, 256]), Frame: 0, Patch: 0

Benchmark Tasks

Task 1: Intra-frame Registration

Register even-line images to odd-line images within each frame, correcting domain shift and geometric misalignment from bidirectional scanning.

Task 2: Inter-frame Temporal Registration

Align the same spatial patch across different time frames. Samples sharing the same patch_idx but with different frame_idx values form a temporal sequence for registration and motion correction.

Citation

If you use this dataset, please cite:

@misc{tianyan_zhang_2026,
  author    = {Tianyan Zhang and Chengliu Yan and Xiangzhi Lan},
  title     = {OR-PAM-Reg-Temporal-26K (Revision f8bffec)},
  year      = {2026},
  url       = {https://huggingface.co/datasets/chengliuyan/OR-PAM-Reg-Temporal-26K},
  doi       = {10.57967/hf/7723},
  publisher = {Hugging Face}
}

License

This dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).