TSlam-dataset

The current repository contains the evaluation dataset for TSlam (@IBOIS, EPFL 2023). TSlam is a SLAM algorithm developed at IBOIS (Laboratory for Timber Construction) for fabrication scenarios involving cutting and drilling operations.<br> The algorithm's accuracy and precision have been gauged using the current dataset and an optitrack system to collect ground-truth data. The dataset is shared in open-access with the goal of promoting the reproducibility of our results and the pipeline for the evaluation can be found in https://github.com/ibois-epfl/TSlam/tree/main/eval . The structure of the dataset is as follow: ```<br> ├───01_camera_recordings<br> │ ├───01_fabrication-20230421-170830<br> │ ├───01_manual_mark_20230419-110238<br> │ ├───01_mapping_recording-20230421-133017<br> │ └───01_tag_mark_20230419-155416<br> ├───01_label_sequence<br> ├───01_opti_recordings<br> │ └───__raw<br> ├───01_pcd<br> │ ├───01_aligned_pcd<br> │ └───__raw<br> └───01_tslam_map_files<br> ``` Where, - `01_camera_recordings`: contains the video from the fabrication, marking, tags sticking and mapping;<br> - `01_label_sequence`: human-noted labels of the start and end frame of all operations in a given fabrication video;<br> - `01_opti_recordings`: the data obtained from the optitrack recordings during fabrication<br> - `01_pcd`: the scans and processed point clouds of the piece before the fabrication<br> - `01_tslam_map_files`: the files produced by TSlam