Aligned Bright-Field and SHG Image Pairs for Registration Method Evaluation

206 aligned Bright-Field and Second-Harmonic Generation image pairs, each 834x834 pixels, for evaluating registration methods. The dataset includes a training set of 40 pairs, two validation sets of 25 and 7 pairs, and a test set of 134 pairs with applied random transformations. It was produced by Kevin W. Eliceiri of the University of Wisconsin–Madison, based on data from a breast carcinoma survival study.

Use Cases

• Benchmarking multimodal image registration algorithms based on the provided aligned Bright-Field and SHG image pairs.
• Training supervised learning models for image alignment using the 40-pair training set with known correspondences.
• Evaluating registration method robustness using the 134-pair test set with known random rotations and translations.
• Analyzing registration error using the provided metadata.csv file containing precise corner coordinates and displacement distances.

Strengths

• Includes 206 image pairs with a clear split for training (40 pairs), validation (32 pairs total), and testing (134 pairs).
• Provides precise ground truth for evaluation via a metadata file containing corner coordinates and mean displacement distances for transformed images.
• Test set transformations are controlled, with random rotations up to +/-30 degrees and translations up to 100 pixels, enabling robustness testing.

Limitations

• Column-level documentation for the primary image data is absent; image semantics and preprocessing must be inferred.
• Last update date is unknown; freshness unverified.
• The dataset's scale is limited to 206 image pairs, which may be small for training deep learning models from scratch.

Provenance

Source

University of Wisconsin–Madison, based on data from the study 'Aligned Collagen Is a Prognostic Signature for Survival in Human Breast Carcinoma'.

Collection Method

Images are derived from Tissue Microarray (TMA) cores, with alignment and transformations applied for registration benchmarking.