Running simulation inside instant-ngp app

[mateuszwyszynski]

Hi,

First of all thanks for this great work and for publishing your code.

I am trying to run the simulations on provided datasets. However, after opening the instant-ngp app and starting the training I constantly see only the first timestamp from the dynamic scene, i.e. there is no change in the object's state. I've already tried the spring and the robot dataset. The spring has constantly the same size and the robot arm is not moving around the scene. How can I obtain results similar to those in the animations on your project page?

I can see that the datasets are divided into timestamps 0, 1, 2, ..., but I couldn't figure out how to move from timestamp 0 to the later ones.

Thanks for your help,
Mateusz

[jc211]

Hi Mateusz,
The frames are updated through the python API at the moment, so you will have to use python particle_scripts/run.py to step through the animation instead of using instant-ngp directly.

[mateuszwyszynski]

Thanks for the reply! I've managed to run the experiments.

I have encountered a few minor issues when trying to run the run.py python script. I dealt with them on my own, but I will describe them here in case someone else falls on the same path:

1. The run.py script is used to run all of the experiments. To run a single experiment I used either run_experiment.py or run_testbed.py
2. When using run.py or run_experiment.py I had to:

• install conda --> the run_experiment.py script uses a system command which relies on conda being installed to invoke run_testbed.py
• change python to python3 inside the command used in the run_test function in the run_experiment.py script
• provide a custom path to the dataset and a custom path to save the results in the experiments.json file

I ended up forking the repo and rewriting the scripts in such a way that conda is no longer required and there is no issue with python command - just to keep things clear for myself. Basically I've changed everything so that the necessary functions are imported from other scripts instead of invoking system commands. But still, the quickest solution is just installing conda and changing python command if needed.