Pix4DMatic: Cluster computing

PIX4Dmatic PIX4Dmatic Feature Request
1

Dear all,
Pix4DMatic come to fill a great hole in the PIX4D ecosystem. However, it always miss cluster and parallels computing. Please do you plan to allow us to use multiple workstations working on a same project using clustering and failover computing?
Please let’s know about this very important request.

Best

2

Dear haitiUser,

Thank you for the feedback and suggestion.

I will ask you a few questions to better understand the context that lead to this request, this will help to understand the goal you would like to reach.

  • May I ask you for what type of projects you would need such cluster and parallel computing? e.g. what is the use case, how many images,…
  • What is the main objective you would like to reach with this? e.g. faster processing, larger data processing,…here I’m trying to understand what part of your work you would like to improve.
    *What kind of hardware would you want to use for this?

We are considering cluster and parallel computing for the future development of our algorithms, but your information on the context will help to prioritize this correctly. This should also enable other people from the community to post their request here.

Related to this we were thinking of an “iterative processing”, e.g. if you fly 10 flights over an area, you could start processing the images of the first flight as soon as the drone lands. When the second flight ends you can add these to the first processing without the need to restart the project from scratch. Maybe this would already solve part of the problem?

Thank you!
Looking forward to hearing from you.

Regards,
Pierangelo

3

Dear @Pierangelo_Rothenbuhler,
Cluster computing is the main missing piece for PiX4D. Please see my response below:

  • May I ask you for what type of projects you would need such cluster and parallel computing? e.g. what is the use case, how many images,… We’re implementing some projects at very high resolution (less than 1cm) for cadastral and crop identification. For example, a 150ha @1cm with a DJI X5 16MP series camera require 25000 pictures. However we have 600Ha and 4000 ha for processing.
  • What is the main objective you would like to reach with this? e.g. faster processing, larger data processing,…here I’m trying to understand what part of your work you would like to improve.Yes: Faster processing of large dataset is the main objective. We can’t spent one week for processing time on a a 450ha project. And if something is wrong, we take another week… With clustering computer, we can scale workstations for project
    *What kind of hardware would you want to use for this?We’re using Dell Workstation and we intend to buy another upgrade with dual 14C Xeon processor (28C total) and 256GB DDR4 RAM with GTX 1080Ti card. So clustering would allow us using two workstations on a single project.

Related to this we were thinking of an “iterative processing”, e.g. if you fly 10 flights over an area, you could start processing the images of the first flight as soon as the drone lands. When the second flight ends you can add these to the first processing without the need to restart the project from scratch. Maybe this would already solve part of the problem?
Cluster computing is a better approach. Right now 10Gbe ethernet is affordable. SSD SAN also. For example in our office we have several servers with Xeon multicores processors, lot of RAM and running on Windows Server. When you support cluster, we can use these processors on a project. This will allow us using existing hardware rather than buying more. Others software such as Open Drone Map and Agisoft Metashape already implement this. However we think that PIX4D is the golden standard. So, we prefer having this feature and improvement in it. Iterative computing is a very good idea, but this require more GCP: lot of for each part rather than if we can process a larger area. Don’t forget, until now, PiX4D doesn’t support PPK…

1 Like
4

Dear @haitiUser,

Thanks for the additional information. Impressive projects you have…your request is duly noted. This might even be something for Pix4Dengine (once this supports parallelization as well).

As for the remark on PPK, we’re planning to add a .csv import possibility to the software, so that you can import geotags for PPK workflows.