MovieLite: A Lightweight Dataset for Movie Genre Classification

The video dataset for classifying movie genres was created based on individual scenes, we adopted a collaborative, human-centric data collection approach. A diverse set of movies in various languages was selected to ensure broad representation across different cultural and cinematic styles. Each movie was segmented into distinct scenes based solely on visual content, without considering audio or dialogue. These scenes were carefully extracted to focus on visual cues such as setting, characters, and actions. Each scene was then independently reviewed by multiple team members. After watching the scenes, each member provided their opinion on the most appropriate genre label, such as action, comedy, drama, or horror, based on the visual content. The genre category that received most votes was assigned as the final label for that scene. In cases of ties or disagreements, group discussions were held to reach a consensus. The final labels reflect the scene type as interpreted visually and were manually assigned by the data collectors based on collective opinion. The resulting dataset of labeled scenes serves as a foundation for training and evaluating models for movie genre classification