GTA-IM 3D Human Scene Interaction Dataset

GTA-IM stands for GTA Indoor Motion, which is a GTA indoor motion dataset collected by the GTA game engine from the Grand Theft Auto (GTA) video game. This dataset has clear 3D human pose and camera position annotations, and has great differences in human appearance, indoor environment, camera view, and human activities.

The dataset has 1 million RGB-D frames with a resolution of 1920×1080 pixels, and the ground truth 3D human pose is 98 joints, covering various actions including sitting, walking, climbing, and opening doors. Each scene contains multiple settings, such as living room, bedroom, and kitchen, to emphasize the interaction between people and scenes.