How do the visualization part work

devendrachaplot / Neural-SLAM

Pytorch code for ICLR-20 Paper "Learning to Explore using Active Neural SLAM"

http://www.cs.cmu.edu/~dchaplot/projects/neural-slam.html

MIT License

761 stars 144 forks source link

How do the visualization part work #28

Closed QieSima closed 3 years ago

QieSima commented 3 years ago

Hello! I am curious about the visualization of your project especially the generation of map with exploring trajectory like the gif in your website. I look through the code and didn't find how the visualization part works. Could you give me some hints? By the way, could you tell how much time did you spend in training and how many GPUs you have used ? Thanks!

devendrachaplot commented 3 years ago

For visualization, just add -v 1 argument to the run command, see instructions here: https://github.com/devendrachaplot/Neural-SLAM#for-evaluation

The visualization code is in this file: https://github.com/devendrachaplot/Neural-SLAM/blob/master/env/habitat/utils/visualizations.py

We trained the model with 72 parallel threads on an 8-GPU system for 10 million frames. The amount of time varies from system to system, depending on the number of GPUs you use and GPU & CPU speeds.