Drone Discussion Presentation

class: center, middle

# Fort Huachuca East Range Reflections: Programs, Compute Resources, and Lessons Learned

## Devin Bayly

Data and Visualization Consultant Research Technologies UArizona

---

# Agenda

1. Personal Background
2. Compute Access
3. Project Experience
4. Questions

---

# Personal Background

* Math and Neuroscience
* Visualization as retrospective
* Web visualization (Mozilla Iodide, early Observable Competitor)
* UA HPC
* **Note: No photogrammetry education!**

---

# RTDV

* Services
    * DCC
    * High Performance Visualization
    * Visualization adjacent support (project collaborations) 
* Past Projects
    * https://rtdatavis.github.io/
    * ["Tell it to the river"](https://www.samaalshaibi.com/exhibitions-1/tell-it-to-the-river-at-maraya)
    * ["Stellarscape"](http://www.stellarscape.org/)
    * [PhytoOracle VR SXSW](https://sxsw.arizona.edu/person/phytooracle-2-vr)

---

# Getting in touch

* vislab-consult@list.arizona.edu
* [Data & Vis Tuesdays 9-11am](https://libcal.library.arizona.edu/event/10712321)
* [Code commons Wednesdays 2-6pm](https://codecommons.net/)

# Support material

* [HPC visualization](https://public.confluence.arizona.edu/display/UAHPC/Visualization)
* [Blender USD Point Cloud support](https://public.confluence.arizona.edu/display/UAHPC/Visualization)
* Lidar processing OSG science gateway (underway)

---
# Compute Access: On Campus

* [HPC](https://public.confluence.arizona.edu/display/UAHPC/Account+Creation)
    * Workshops updates by email
* [Cyverse DE](https://cyverse.org/discovery-environment)
* [Visualization Wall](https://lib.arizona.edu/catalyst/spaces/data-studio)
* [Catalyst Studios](https://lib.arizona.edu/catalyst)
* [Sensor Lab](https://sensorlab.arizona.edu/)

---
# Compute Access: Off Campus Specific Programs

* RMACC 
    * [Alpine CU Boulder]( https://www.colorado.edu/rc/alpine)
    * [Cumulus (Persistent Cloud Compute)](https://www.colorado.edu/rc/resources/CUmulus)
* [OSG](https://osg-htc.org/)
* [Jetstream 2]( https://jetstream-cloud.org/index.html)
    * [exosphere]( https://jetstream2.exosphere.app/exosphere/home)

---
# Compute Access: Off Campus General

* [Campus Champions](https://campuschampions.cyberinfrastructure.org/)
* [ACCESS](https://access-ci.org/)
    * My allocation (demo)
    * Assistance writing Explore Allocation

---
# Fort Huachuca Project: Background

* Supporting students and PI from Digital Humanities
* Reconstructing the east range 
* ~145km^2 ( yikes! )
* Wingtra one (no ppk)
* 1cm resolution
* \>60k images
* Open Drone Map
* Desired outputs 
    * DEM & Ortho
    * Pointcloud (.laz)
    * 3d textured models
    * VR assets
    * OGC 3d Tiles
    * (different departments want different outputs)
* [project repo](https://github.com/DevinBayly/FT_photogrammetry)
    * includes qgis and odm scripts
    * under development

---
# Fort Huachuca Project: Workflows

* WebODM on HPC (vanilla)
    * Converted Docker Compose to Singularity instances
    * ElGato 16 core 12 hour instance, 200-800 images success
    * pro:
        * using OOD remote desktop can teach students to use
    * con:
        * manual submission will not scale to image numbers we have

---
# Fort Huachuca Project: Workflows

* ODM Split Merge Cluster (sm-cluster)
    * [support doc](https://docs.opendronemap.org/large/)
    * Setup 8 worker nodes
    * put their configuration info in cluster-odm
    * launch cluster-odm
    * launch odm container with `--split 200` `--sm-cluster http://localhost:3000`
    * pro:
        * can  point at entire folder of images and start processing programmatically
    * con:
        * Image products depend on flight pattern (long and narrow feature artifacts more than square tile)
        * unpredictable OOMs and allocation time limits exceeded
        * no all image products can be merged (no 3d)
---
# Fort Huachuca Project: Workflows

* QGIS group segmentation SLURM HPC ODM
    * establish tiled overlapping groups of images from the different flights
    * export a large JSON with those groups
    * Implement a python script that launches HPC tasks (12-16 hours) aimed at processing each group separately
    * leverage .las information for aligning the 3d models
    * pro:
        * reliably completes without OOM, and/or timeouts
    * con: 
        * hard on the presentation end of the pipeline
---
# Fort Huachuca Project: Difficulties

* Missing files
    * Experiencing a lot of problems filling in sections of map
* Storage issues
    * 10TB runs out fast
* ODM requires a lot of parameter education
    * missing manual helps with this

---
# Closing thoughts

* ODM is a great program with a very active community, recommend experimenting with it
* Reach out to me for any assistance with getting compute access 
* Visualization and DCC project support
* Acknowledgment
    * Tyson Swetnam Jeff Gillan
    * Support from Piero Toffanin