all_results.zip

This NumPy data file contains a structured collection of experimental results from evaluating various prompt composition methods. The dataset includes CLIP similarity scores for 1,152 generated images, measuring their semantic alignment with source prompts across six distinct categories (Simple, Natural, Urban, Abstract, Emotional, Technical). For each composition instance, the file stores detailed metrics including CLIP scores, semantic preservation values, generated prompts, corresponding images, and the specific composition methods used. The data structure follows a dictionary format with keys for 'clip_scores', 'categories', 'semantic_metrics', 'prompts', 'images', and 'methods', enabling comprehensive analysis of the effectiveness of different prompt composition strategies. The image data is stored as a numpy array with dimensions (1152 x H x W x C), while other metrics are preserved in their respective native formats for easy processing and evaluation.