cvis-tmu/spar7m-rl

SPAR7M-RL数据集是从SPAR-7M JSONL源转换而来，具有固定的测试分割。它包含多种问题类型，如外观顺序、相机运动推断、深度预测（包括遮挡和对象间场景）、距离推断（涉及中心点、遮挡和对象间关系）、对象计数、对象空间关系、位置匹配、房间大小估计、空间想象和视图变化推断等。这些类型分为fill（填空）和select（选择）任务，样本数量在0到5000之间不等，覆盖了空间推理和视觉理解领域的多样化任务。数据集旨在支持强化学习或相关NLP/计算机视觉研究，提供结构化的问题-答案对。

Dataset converted from SPAR-7M JSONL sources with a fixed test split. It includes various question types such as appearance_order, camera_motion_infer, depth_prediction (with occlusion and object-to-object scenarios), distance_infer (involving center points, occlusion, and object relationships), obj_count, obj_spatial_relation, position_matching, room_size, spatial_imagination, and view_change_infer. These types are divided into fill and select tasks, with sample counts ranging from 0 to 5000, covering diverse tasks in spatial reasoning and visual understanding. The dataset is designed to support reinforcement learning or related NLP/computer vision research, providing structured question-answer pairs.