Brain MRI Images for DPO Training with 15,320 Positive Samples

Brain MRI images used for a matched comparison between Gaussian-noise and SD-generated negative images in Direct Preference Optimization training. The repository includes 15,320 positive training images and references unique lesion masks, alongside a held-out Figshare segmentation test set. The dataset was authored by KRMayD and last updated on Hugging Face in May 2026.

Use Cases

• Training Direct Preference Optimization models for medical image generation based on the described comparison between Gaussian and SD-generated negatives.
• Evaluating brain lesion segmentation algorithms using the held-out Figshare test set mentioned in the description.
• Researching the impact of different negative image generation techniques on DPO training outcomes for brain MRI data.

Strengths

• Includes 15,320 positive training images for DPO model training.
• Provides a held-out test set from Figshare for segmentation evaluation.
• Focuses on a specific, structured comparison between two negative image generation methods.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.

Provenance

Source

Hugging Face dataset authored by KRMayD.

Collection Method

Likely compiled for a specific machine learning experiment comparing DPO training techniques.

Freshness

Last updated 2026-05-24 10:18:40; freshness should be verified.