Appendix
Last updated
Last updated
This GitBook manual is connected to a GitHub repository that contains all the files used to recreate this project.
It might be useful to see a video tutorial going through the steps outlined in here for further clarity. This tutorial is available here:
The Blender file used in the tutorial is available to download here.
If you plan to render on Adroit-Vis, you might be tempted to parallelize your program, by running multiple instances of the same command (at the same time), each rendering a different part of the sequence. However, when we render out a sequence, Adroit automatically splits up the work among its two GPU nodes. This means that there's no point in parallelizing, or running multiple instances, as it'll just "overload" the GPUs by giving already running GPUs more work, ultimately slowing down the render.
Here's a look at both Adroit-Vis GPUs as we render out our scene. You can see that both GPUs are using our Blender process, without us having to parallelize:
This output can be viewed by opening a new Adroit-Vis terminal and running nvidia-smi while the rendering process is ocurring.
Here are some notes about the final video:
The final rendered output in Blender was then editing using Adobe Premiere Pro
The voiceover was recorded using Voice Memos and processed using Adobe Podcast
The reference for some of the statistics about the percentage of dead trees within Hunts Point comes from manipulating the data as described in this R script in the repo
The source for the statistic about Black New Yorkers dying at twice the rate as white ones comes from NYC Environment and Health
The source for the information about trees and runoff comes from the EPA
In the original Blender file in GitHub, there are a few minor discrepancies between what's within that file and what's explained here. In general, you should tend to follow the directions specified within this book rather than copying any errors that exist in the original file:
The alignment of the buildings to the rest of the BlenderGIS imports was done manually rather than using the CRS and alignment provided in Blosm
I only discovered the very elusive solution to Blosm projections after the final render was complete
There are several unneeded keyframes for the render visibility keying - they make no difference in the final product
The keyframes are structured as if render visibility can be interpolated between frames, even though it is binary (i.e. no need to have keyframes that are adjacent to each other to switch between on/off)
This reference book was developed in 2024 by Nikhil Chinchalkar under the guidance of Carolina Roe-Raymond and Eliot Feibush for Princeton University's Research Computing department.
