Datasets
Standard Dataset
Users' Trajectory Training Dataset
- Citation Author(s):
-
JianxinSun
- Submitted by:
- Jianxin Sun
- Last updated:
- Mon, 06/10/2024 - 16:11
- DOI:
- 10.21227/7qfn-t458
- Data Format:
- Research Article Link:
- License:
90 Views- Categories:
- Keywords:
Abstract
The training trajectory datasets are collected from real users when exploring the volume dataset on our interactive 3D visualization framework. The format of the training dataset collected is trajectories of POVs in the Cartesian space. Multiple volume datasets with distinct spatial features and transfer functions are used to collect comprehensive training datasets of trajectories. The initial point is randomly selected for each user. Collected training trajectories are cleaned by removing POV outliers due to users' misoperations to improve uniformity. The distribution of the training trajectories is balanced for they span the whole domain containing the volume rather than only focusing on a small region of interest. The visualization task is to locate the regions of interest to each user through exploration.
The datasets used to train the model are sequence and label pairs cut from the 50 training trajectories (340 Point of Views per trajectory) by applying a sliding window. Since a window size of 3 is used as the input sequence length for training the RmdnCache predictive model, there are in total 16850 sequence and label pairs for training. During the training process, the training and validation datasets can be split from all the sequence and label pairs.
Structure:
- Each training trajectory is a .txt ASCII file.
- The index of the trajectory is x of the x.txt file name.
- The first line of each .txt file is the location header in Cartesian space with order of x, y and z.
- Each following line is the spatial location of a Point of View (POV).
- The line order in the .txt file follows the same order as the POVs of the user's trajectory.
