A Few Rookie Questions

Hey everyone,

Relatively new to droning, brand new to mapping and WebODM (local install), but fascinated by both. I’m going to try and ask a few questions in this post; have searched the forum and not found any clear answers.

  1. RTK GPS - Can anyone point me in a direction towards an RTK solution for GCP generation? This will be the only use for the unit, so I’m not looking to break the bank. I intend on doing some agricultural mapping, and 3d construction progress site mapping to import into ACAD. I’ve generated a few maps, but without GCPs swome of my results are very skewed and squirrely.

  2. Dedicated headless processing box - What kind of OS, Memory capacity, and processor capability would be recommended for a headless processing system? I’m a windows guru and know my way around linux distros and package installs. I just can’t have my desktop or laptop tied up for 23 hr (and running) processing sessions. Any advice is welcome. Again, not looking to break the bank.

  3. It seems evident to me that between the WebODM processing and final deliverables there are quite a few steps utilizing QGIS, WebMesh, and other tools. Can anyone point me in a direction towards a workflow from drone to ACAD?

Thanks in advance for anyone willing to help me out.

2 Likes

Welcome!

  1. We have some incredibly knowlegeable folks here. I’d start with a search about RTK GPS and picking through some of the threads. Really excellent information:
    Search results for 'rtk gps' - OpenDroneMap Community

  2. Linux with Docker will likely be recommended. I personally prefer Alpine Linux and use it for all my Linux machines because I’ve found it to be easy to work with, lightning fast to install/deploy, and bullet-proof for package and OS upgrades unlike every other distro I’ve tried. CPU, RAM, etc will be highly dependent upon your anticipated job sizes in terms of image count, image resolution, and your desired final product quality. Try to flesh these details out.

  3. Not necessarily. Some folks certainly do use these other tools and workflows, and others do not. It all depends upon what you need/want from your final deliverables.

1 Like

With all due respect, I appreciate your time typing that response, but nothing in your post resembled an answer to any of my questions.

Any help, anyone?

1 Like

For points 2 and 3 I would go the same way as Saijin_Naib. All depend on what you are using ODM for.
Can you provide more info on your projects and purpose ?

  • Area of survey
  • Wanted GSD of orthophoto
  • If already know via existing dataset, number of pictures
  • Output wanted : point cloud ? orthophoto ? dem ?

Some processing can take minutes where others will take a day as you said.

2 Likes
  1. Here’s a similar thread with some great hardware/software recs:
    PPK question - #4 by daniel.dobosi
  2. Knowing your parameters is not optional. Do I reommend to you a server with 768GB of RAM and 128 CPU cores if you’re only making Tasks of 500 images? Please provide more context on your expected parameters so folks can actually help you size appropriate class hardware for your anticipated needs.
  3. Again, this all depends on what your output expectations are. You can take the DEM and Orthophoto products from WebODM straight into AutoCAD, or you can post-process them to warp, reproject, clip, etc if you need to. What is it that you need to accomplish in this stage?
2 Likes

Thanks for that response, that’s good info there. On #2., what I am dealing with now is image sets of 300-450 images. I’m hitting the Out of Memory error pretty frequently every time I try with ultra settings. I’m fooling around with RAM and Swap disk size, which seems to be having an effect, but still not getting what I want.

Current machine is Windows 11, AMD Ryzen 5 1400 Quad-Core Processor 3.20 GHz, 25GB RAM.

Current Command Line params is auto-boundary: true, dsm: true, dtm: true, feature-quality: ultra, gps-accuracy: 1, matcher-neighbors: 8, max-concurrency: 2, pc-quality: ultra

1 Like
  1. What resolution are the images (typically)? 12MP? 16MP? This can have an outsized effect when using ultra parameters and no resizing.

I would try setting your Pagefile to a static 128GB if you have the room for it. The pagefile can dynamically expand, but if there’s any latency in doing so, it can cause an Out-Of-Memory despite the potential full size of the pagefile being able to accomodate the spike.

Do you have the ability to upgrade the RAM on that machine? 64GB with 64GB pagefile should work well for that size Task unless the images are gigantic.

1 Like

Image resolution is 4000 x 2250. A question I’d like to ask here is, should I use 4:3 or 16:9 for my image acquisition? Does this make a difference?

I’m using a DJI mini 2 right now, which I know is not optimal but the success I had with smaller image sets did show me that the outputs are sufficient for what I want to do. The DJI mini 2 has a 12MP camera. Picture sizes are around 4MB each.

For WebODM I am using WSL2 right now, with .wslconfig setting memory at memory=16GB , processors=2, swap=120GB. Should I swicth to the Hyper-V engine for this?

1 Like

Flight planners assume native aspect ratio, so 3:2 or 4:3. I would use that at the full 12MP. And yes, it makes a massive difference. Your overlap calculations from flight planners will be way off for instance, making reconstruction more difficult (or in some scenes, impossible).

No need to change, no, WSL2 is the preferred/recommended engine for Docker. Are you actively using the Host OS during processing? You might be able to get away with bumping the memory cap in the .wslconfig to 20GB and leaving Windows with 5GB (so long as it is backed by a pagefile and you’re not actively using it, this should be fine).

With 16GB, I wouldn’t recommend going over “high” for your quality parameters (and even then, medium might be more reliable).

Does this machine have upgradable RAM? Is that in your budget?

1 Like

OK on the native aspect ratio. I did suspect that was a problem. Since you brought up overlap, I am using 83% front and side. Can you give me a recommendation for that?

Unfortunately, I am active on the host machine about 50% of the time; using my work PC. The 16GB is taking a toll on the OS, LOL, but all I need is Outlook, Word, and Excel.

This machine is capable of max 32GB, and yes, I will be making this change as soon as I get the RAM modules.

Is the large swap file helpful? SHould I increase the size of BOTH the windows pagefile as well as Docker/managed swap file?

1 Like
  1. Overlap should go as high as your platform allows, I think. It only costs you storage and (a bit extra) processing time, but can help a lot in difficult scenes. Sidelap is what costs you airtime as tighter sidelap means more flight transects. We recommend 65% at a minimum for 2D mapping and 85% for 3D mapping.

  2. Ahh, yep, I understand. I’m in the same boat. I would suggest you think about budgeting for a build that can support 64GB+ RAM with your projected Task size/settings, and price it against using WebODM Lightning (or a self-host of WebODM on Azure, AWS, Google Cloud, etc).

  3. Yes, it is helpful, but it is painfully slow and can still lead to Out-Of-Memory if the drive hosting it is too slow to respond. If you’re running the Host OS really slim on physical memory, having a pagefile to back it up will help with the Host OS’ stability, but OpenDroneMap within Docker/WSL2 won’t directly touch it. You’ll be best served by having a sufficiently sized swapfile for Docker/WSL2 as well as a pagefile for Windows to manage its memory with.

1 Like

Roger that. Thank you so much, you’ve been very helpful. Sorry for my snarky initial response. I will keep you posted with where I go from here.

2 Likes

Keep us apprised of what you’re up to! If you have something cool, try #the-showroom to show it off!

And no worries :slight_smile: You don’t know me so I may have come off dismissive where I was trying to lead you to discover a bit and ask follow-ups that we could build upon. Something for me to work on, for sure.

2 Likes

No problem, you came back and poured the information on strong, LOL.

I’m wrapping up in the office today, but did fly the neighborhood with the 4:3 AR, and successfully processed 613 images in 04:28:19 with the cmd-line being as follows:
auto-boundary: true, dsm: true, dtm: true, gps-accuracy: 1, matcher-neighbors: 8, max-concurrency: 2, pc-quality: high

I’m currently running the same cmd but changed to ultra. I’d like to show you the ortho and 3d models tomorrow as they came out, if you don’t mind giving a quick assessment of what I ended up with. I really appreciate your help. I’ve only been flying drones since January, but got my part 107 in Feb, and I’m trying to marry this into my full time job as a federal contractor construction PM consultant.

1 Like

Awesome! Glad it reconstructed.

Sure, share it here and I’ll take a poke when I can. I’m sure others can offer valuable feedback as well, so be on the lookout!

1 Like

I would like to add the you should be sure of your objective.
This is very likely that you do not need ultra setting depending on result quality you want.

2 Likes

Here’s a summary of a test I ran with some super high quality settings on that same image size, in case helpful…

WebODM Settings:
auto-boundary: true, dsm: true, feature-quality: ultra, matcher-neighbors: 40, mesh-octree-depth: 12, mesh-size: 300000, min-num-features: 64000, orthophoto-resolution: 1, pc-geometric: true, pc-quality: ultra, rerun-from: dataset, resize-to: -1

285 images at 4000 x 2250px resolution
Processing time: 4h10m
Max RAM consumption: 51GB
Max hard disk consumption: 42GB

And for comparison, with higher resolution images & bigger quantity…
428 images at 5472 x 3648px resolution
Processing time: 17h51m
Max RAM consumption: 148GB
Max hard disk consumption: 154GB

2 Likes

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.