Kurosawa Shot-Level Dataset: 30 Films with Shot Boundaries, Color, and Face-Derived Shot Scale Features

If you use this dataset, please cite our official data paper published in JOHD: Wu, X. (2026). Kurosawa Shot-Level Dataset: Shot Boundaries, Color, and Face-Derived Shot-Scale Features for 30 Films. Journal of Open Humanities Data, 12(1), 30. https://doi.org/10.5334/johd.507 This record releases a shot-level dataset for a film-level census of 30 feature films directed by Akira Kurosawa. The dataset provides automatically extracted shot boundaries and per-shot visual descriptors designed for quantitative film-style analysis. Contents Shot boundaries: shot start time, end time, and duration (seconds), including film-level summary statistics (e.g., raw ASL and corrected ASL). Shots shorter than 0.5 s are excluded from downstream summaries to reduce noise from extremely short segments and decoding artifacts. Within-shot visual descriptors (deterministic sampling): for each retained shot, five frames are sampled at fixed relative positions (1/6, 2/6, 3/6, 4/6, and 5/6 of shot duration). The same sampled frames are reused for all feature types. Color features: HSV saturation and CIELAB-derived brightness and warmth (computed on a central ROI and aggregated at the shot level by averaging over the five frames). Face/shot-scale features: face counts and a face-size proxy (maximum face bounding-box height divided by frame height) computed per sampled frame. A shot is coded as face-present if faces are detected in at least 3/5 sampled frames (coverage ≥ 0.6). Shot scale is assigned using predefined thresholds on the median of the per-frame maximum face-size proxy; otherwise, the shot is coded as No_Face/Other. Files kurosawa_shot_level_data.csv — consolidated shot-level table for all 30 films (CSV, UTF-8). Kurosawa30.zip — per-film shot-level CSV files (one CSV per film). generate_kurosawa_dataset.py — analysis script (Python 3.10) used to generate the CSV outputs. Reproducibility and reuseThe included analysis script documents the extraction and aggregation procedures used for this release. The dataset contains derived numerical measurements only and does not include any film video, audio, or frame images. It is intended for research and educational reuse in computational film studies and related quantitative media analysis.