in-house gameplay event detection dataset
收藏资源简介:
本研究构建了一个内部游戏事件检测数据集,该数据集包含超过110 GB的视频数据,其中包含五款游戏中的七种事件类型(不包括背景事件)。数据集由AMD的测试人员通过手动标注视频片段中的事件时间间隔而创建。数据集的总事件数为7153,其中包含4408个个体事件和2745个背景事件。该数据集旨在通过使用X-CLIP模型进行微调,实现跨多个游戏的泛化,从而无需针对每个游戏进行工程化。数据集的创建过程耗时且迭代,需要确保收集的事件数据不会导致数据偏差。数据集在游戏社区中具有广泛的应用前景,特别是在第一人称射击(FPS)游戏中,可用于自动生成吸引人的游戏精彩集锦,从而提高游戏玩家的社交互动和观众参与度。
This study constructs an internal game event detection dataset, which contains over 110 GB of video data covering seven event types (excluding background events) across five games. The dataset was created by AMD testers via manual annotation of event time intervals within video clips. The total number of events in the dataset is 7153, including 4408 individual events and 2745 background events. This dataset is designed to achieve cross-game generalization through fine-tuning the X-CLIP model, eliminating the need for per-game engineering customization. The dataset creation process is time-consuming and iterative, requiring guarantees that the collected event data will not introduce data bias. The dataset has broad application potential in the gaming community, particularly in First-Person Shooter (FPS) games, where it can be used to automatically generate captivating game highlight reels, thereby enhancing social interaction among players and audience engagement.

- 1Gameplay Highlights GenerationAMD · 2025年



