TCN Model for Trajectory Inpainting with Symmetric Dilation
Researchers propose a Temporal Convolutional Network (TCN) with symmetric dilation to reconstruct missing segments in trajectory data, a task known as trajectory inpainting. Unlike forecasting-oriented architectures, the model relaxes the causality constraint to use both past and future observations. A composite loss function combines weighted mean squared error, boundary-continuity penalties, and smoothness regularization. The model was trained on a synthetic dataset of 1,000 training, 200 validation, and 300 test two-dimensional trajectories with randomly placed 20% masked segments, achieving strong R², MSE, and MAE metrics.
Key facts
- Trajectory inpainting reconstructs contiguous missing segments from observed context.
- TCN with symmetric dilation relaxes standard causality constraint.
- Composite loss includes weighted MSE, boundary-continuity penalties, and smoothness regularizer.
- Dataset: 1,000 train, 200 validation, 300 test two-dimensional trajectories.
- 20% of each trajectory is randomly masked.
- Model achieves good R², MSE, and MAE metrics.
Entities
—