This dataset is part of the work presented in the paper AeSlides: Incentivizing Aesthetic Layout in LLM-Based Slide Generation via Verifiable Rewards.
AeSlides is a reinforcement learning framework with verifiable rewards for aesthetic layout supervision in slide generation. This benchmark focuses on quantifying slide layout quality through verifiable metrics like aspect ratio compliance, whitespace reduction, and visual balance.
GitHub Repository:… See the full description on the dataset page:
https://huggingface.co/datasets/ympan/aeslides-reward-bench.