InFlux
收藏资源简介:
InFlux是一个包含动态内参视频的基准数据集,为每个视频帧提供真实的内参标注。该数据集由386个高分辨率室内和室外视频组成,总计超过143,000个标注帧。这些视频展示了动态相机内参变化、多样化的场景和相机运动。为了确保每帧内参的准确性,我们构建了一个全面的校准实验查找表,并扩展了Kalibr工具箱以提升其准确性和鲁棒性。InFlux旨在为研究提供真实世界的动态内参视频,以便评估现有方法在预测相机内参方面的性能。
InFlux is a benchmark dataset containing videos with dynamic camera intrinsics, which provides ground-truth intrinsic parameter annotations for each video frame. This dataset consists of 386 high-resolution indoor and outdoor videos, totaling over 143,000 annotated frames. These videos showcase dynamic camera intrinsic variations, diverse scenarios, and various camera motions. To ensure the accuracy of per-frame intrinsic parameters, we constructed a comprehensive lookup table for calibration experiments and extended the Kalibr toolbox to improve its accuracy and robustness. InFlux aims to provide real-world videos with dynamic camera intrinsics for research, so as to evaluate the performance of existing methods in camera intrinsic parameter prediction.




