Text-2-Image DPO Human Preferences: 80,000 Trust-Weighted Pairwise Judgments

80,000 trust-weighted pairwise judgments from calibrated annotators compare AI-generated images on prompt alignment and overall preference. The dataset was built on the Datapoint annotation platform for collecting high-quality human preference data at scale. It was authored by datapointai and last updated on March 30, 2026.

Use Cases

• Fine-tuning text-to-image models via Direct Preference Optimization based on human preference judgments.
• Training reward models for reinforcement learning from human feedback (RLHF) in image generation.
• Benchmarking the alignment of generative models with human preferences on prompt fidelity.
• Studying the correlation between prompt alignment and overall aesthetic preference in generated images.

Strengths

• 80,000 pairwise judgments provide a substantial basis for model training.
• Data is quality-controlled and uses trust-weighted judgments from calibrated annotators.
• Built on a purpose-built annotation platform designed for high-quality preference collection.

Limitations

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

Provenance

Source

datapointai via Hugging Face

Collection Method

Collected via the Datapoint annotation platform using calibrated human annotators.

Freshness

Last updated 2026-03-30 06:37:47; freshness should be verified.