JTAP: Joint Physical Tracking and Prediction

jtap is a Python library for probabilistic psychophysical modeling and inference of 2D Mental Physics. Project page: https://arijit-dasgupta.github.io/jtap/

Installation

Requirements

To run jtap, you must use a machine with at least a single NVIDIA GPU (minimum 24GB Memory) and have CUDA 12 (with the appropriate driver version).

Setup

Follow the following code block to get jtap running on your machine. You will need uv installed.

uv sync

Reproducing CogSci 2025 Results

To reproduce the results from the paper, run

python run_jtap_cogsci2025.py

Since JTAP is probabilistic, results might vary slightly, but draw the same conclusions as the paper. Running the experiment to reproduce the paper results make take 10-15 minutes.

Three pdfs will be generated when running the experiment, which are identically formatted to the figures shown in the paper.