sansangela / mlprediction

[WIP] ML Prediction: Trajectory Prediction

Table of Contents

• Getting Started
• Directory Structure
• To-Do

Getting Started

To run the lstm-model (Work In Progress), run the following command:

pip3 install -r third_party/requirements.txt
python3 social_lstm/main.py

Directory Structure

• data/: Directory for storing dataset files. Referenced from https://github.com/fjhheras/social-lstm-tf.
  • Raw data files are stored in directories eth/ and ucy.
  • preprocessed_data/: Directory for storing preprocessed data files.
• [WIP] social_lstm/: Directory containing Python scripts for social-lstm model.
  • src/: Directory containing Python helper functions and classes for social-lstm model.
  • main.py: Main script for running social-lstm model.

To-Do

The model is currently under devlopment.

• 1. Dataloader: train/val data split (currently only training data).
• 2. Dataloader: all datasets (currently only one dataset for debugging purpose).
• 3. Social-LSTM Model: implement occupancy grid and social embedding layer (currently using a placeholder replicating the input embedding layer).
• 4. Training: implement log likelyhood loss in closed-form (currently using MSE).
• 4. Envrionment: set up Bazil environment (having problem setting up tensorflow).