Pedestrian Trajectory Generation and Visualization

This project trains a UNet1D model on the pedestrian dataset to generate new samples of pedestrian trajectories. The generated samples are saved to a CSV file, and the trajectories of the pedestrians from the generated dataset are visualized using Matplotlib.

Requirements

• Python
• PyTorch
• Matplotlib
• Numpy
• Pandas

Project Structure

• train.py: Contains the script for training the model on the pedestrian dataset.
• sampling.py: Contains the function to generate new samples using the trained model.
• utils.py: Contains utility functions for visualizing and saving the generated samples.
• models/unet.py: Contains the UNet1D model used for generating the samples.
• dataset/plot_dataset.py: Contains the script for plotting the trajectories of the pedestrians from the generated dataset.

Usage

1. Train the model and save the weights to a .pth file, which is saved in the Checkpoints directory, and will be used for generating new samples.
2. Run sampling.py to generate new samples using the trained model. The generated samples are saved to a CSV file in the 'Generated_Positions_Data' directory. The generated samples are saved in the format 'generated_pedestrians_epoch_{}.csv', where {} is the epoch number.
3. Run dataset/plot_dataset.py to visualize the generated samples in the CSV file. The script will plot the trajectories of the pedestrians from the generated dataset, and save the plots to the 'Images' directory.

Results

• The following images show the nomalized trajectories of the pedestrians from the generated dataset at different epochs.

- The following images show the trajectories of the original pedestrians from the dataset.

- The following images show the trajectories of the pedestrians from the generated dataset at different epochs.