Messages

1

General / Aligning photos of an existing project using "Estimated Preselect"

« on: December 30, 2022, 06:45:45 AM »

I was reading through the manual (ver 1. about the "Accuracy" setting for Alignment.

I have a project on a large shipwreck that involves multiple dives. I basically use incremental alignment with the new photos and have some overlap of photos with the previous dives.

What basically happens is that the new photos have a "tight" alignment with each other and a "loose" alignment with the photos from the previous dive that overlap.

This results sometimes in areas where the two sets of photos are close to being aligned, but not exact and there is a little bit of a "ghosting" effect.

I'm generally using Medium alignment accuracy and generic preselect.

With all that...

In reading the manual, I noticed this under "Reference preselection"

"if the alignment operation has been already completed for the project, the estimated camera locations will be considered when the Align Photos procedure is run again with the Estimated preselection selected."

If I didn't want to start from scratch, could I use this option to increase the accuracy?

If so, I'm assuming I would NOT want to check "reset current alignment" but how exactly would I go about doing it? Would I reset alignment of all photos and then select Align Photos and check the Estimated Preselect?

Or would I just run Alignment with High and Estimated Preselect?

If somebody could clarify, that would be great!

Happy Holidays all,

- brett

2

General / Re: Depth maps after incremental photo alignment

« on: December 12, 2022, 08:36:39 PM »

Okay, thanks, understood.

Any thoughts on the other question?

If I start from scratch and re-generate all depth maps, what reasons would cause there to be fewer depth maps than aligned photos?

- brett

3

General / Re: Depth maps after incremental photo alignment

« on: December 12, 2022, 07:40:10 PM »

Thanks, two follow-on questions:

1) If I un-check that option AND I already have some of the maps done, should it just process depth maps for the "new" aligned photos?

2) I have one project where I have 3,404 photos aligned but only 3,377 depth maps were created. What would cause a depth map to not be created for a photo that is aligned? I re-checked that the region included all photos.

- brett

4

General / Depth maps after incremental photo alignment

« on: December 12, 2022, 06:43:18 PM »

I'm running into some behavior that I think is unexpected. I'm using v 1.8 of the Standard version.

Let's say I align 100 photos and build a mesh which creates 100 depth maps.

Then I align an incremental 20 photos. If I then try to build a "new" model, it doesn't create the depth maps for the additional 20 photos or include them in a mesh.

I have Reuse depth maps enabled AND I don't change any of the parameters (e.g., quality, filtering, etc.).

The only workaround I have found is to basically delete all the existing depth maps and start over. If it was just 100 photos, I wouldn't really care, but it is 1000s of photos.

What am I doing wrong?


Thanks,

- brett

5

General / Re: Question on Use CPU when performing GPU accelerated processing

« on: December 02, 2022, 06:33:42 AM »

PS - I know that the manual says to disable this setting when there is a discrete GPU, but I've also seen advice to the contrary on the forum.

- brett

6

General / Question on Use CPU when performing GPU accelerated processing

« on: December 02, 2022, 06:30:00 AM »

I've seen multiple posts on this and I can't seem to find a "best practices" guide on whether to check it.

For example, should I check it if I have a dedicated GPU?

What about if I have an "integrated" CPU in something like the Apple M1 chips?

Should I select it for certain operations and not others?

I've read the manual a couple times and researched some of the forum posts, but I can't find a definitive guide on this.

Thanks,

- brett

7

General / Reduce Overlap - What is considered a "rough" mesh model?

« on: December 02, 2022, 06:26:41 AM »

I've used the Reduce Overlap function in the past but I have a question about what is meant by a "rough" model?

I have generally built a mesh from depth maps but with Quality = Low and Face Count = Low.

Would it be better to build from the Tie Points? I'm assuming it would be much faster but what do you sacrifice? Accuracy?

Is there any "best practice" for what constitutes a "Rough Mesh Model" as mentioned in the user manual?

Thanks,

- brett

8

Feature Requests / Re: Reduce Overlap taking into account cameras estimated quality

« on: December 02, 2022, 06:17:39 AM »

I've been researching "Reduce Overlap" and agree that this would be a great feature.

9

General / Re: Building a mesh from Dense Cloud vs Depth Maps vs Tie Points

« on: December 02, 2022, 02:08:49 AM »

Thanks for your response.

So, if for example, I wanted to have somebody build a 3D model of just an airplane and not the seabed, I would classify the sand one way and the airplane another and then build just the airplane?

Does that make sense?

Thanks,

- brett

10

General / Re: Building a mesh from Dense Cloud vs Depth Maps vs Tie Points

« on: November 30, 2022, 07:40:54 PM »

Thanks. What would be the benefit of generating a dense cloud then? I guess you could do certain editing functions before you build a mesh?

- brett

11

General / Building a mesh from Dense Cloud vs Depth Maps vs Tie Points

« on: November 27, 2022, 02:35:14 AM »

Hi all,

I've seen a few posts about the ability to build a mesh from Depth Maps directly instead of building a Dense Cloud first. I also noticed an option to build from Tie Points.

What are the practical differences in these approaches?

My use case is to build solid models with textures of underwater wrecks and post them to Sketchfab.

I think the "pro" of building from Depth Maps instead of a Dense Cloud is that you would skip the time to generate a Dense Cloud -- but does it increase the time to build a Mesh?

- brett

12

General / Re: Sparse cloud not updated after incremental alignment

« on: November 26, 2022, 06:37:05 AM »

6000 images is not so large dataset to align - what settings are you using and what format/size of images are being used?

Hi Simon,

They are "out-of-camera" JPG from a Sony a6400 so they are 16MB or so each (dimensions 6000px x 4000px with a resolution of 350dpi x 350dpi).

My settings are Key Point Limit of 40,000 and a Tie Point Limit of 10,000. Generic preselection = No, Reference preselection = No, Exclude stationary tie points = Yes, Guided image matching = No, Adaptive camera model fitting = No.


Thanks for any tips or guidance.

- brett

13

General / Re: Sparse cloud not updated after incremental alignment

« on: November 21, 2022, 03:43:28 AM »

A quick update on this. I basically figured out what was going on. After "Dive 4" all the photos had aligned; however, a whole slew of them were aligned in a "rats nest" of photos.

Basically, it was a "butterfly effect" where two pictures aligned but the angle was off and then it spiraled in on itself. The photos for the new dive aligned with the photos in that rats nest and so it just got worse.

I finally figured it out and corrected it but it was painful and I learned my lesson. Always do a sanity check to see if things are correct on the sparse model even if they are aligned.

- brett

14

General / Sparse cloud not updated after incremental alignment

« on: November 19, 2022, 06:00:27 PM »

Hi,

I have a large wreck project I'm working on in phases. I made my fifth dive yesterday and took another set of photos.

I made a low accuracy alignment of JUST the photos from the 5th dive and it worked with 1486 cameras out of 1486 getting aligned  (I've attached a screenshot Dive5-Only) with the workspace.

For my next step, I opened up the existing model that had the results of the previous four dives aligned (4951 cameras out of 4955). I've attached a screenshot Dive4 with that workspace.

You can see the similar areas on the starboard side bow and some of the bridge area.

I then added the photos from Dive 5 into the existing photos from Dive 4->1 and ran a Medium accuracy alignment which took about 6 hours (Note that I had used Tools->Reduce Overlap with 5 cameras per point in order to expedite the alignment of the new photos.

The result was that 6418 photos out of 6441 aligned, BUT the sparse cloud did not reflect the addition of the new photos. I've attached a third screenshot named Dive5-combined.

As far as I can tell, it doesn't include any of the new photos. I've tried saving, closing, re-opening, etc. all to no avail. Given the amount of processing time it would take to re-align all 6441 photos (2+ days I think), I wanted to check in here first.

Any ideas?

Thanks,

- brett

15

General / Re: Performance on Mac devices

« on: November 11, 2022, 10:12:11 PM »

If you are not using some preselection option, then each image is compared against each other image during matching points and 40,000 key points is overkill value. For 5,000 images it means 5,000x5,000x40,000 point to compare...huge number.
Try set 10,000 key points limit and 4000 or even less(3,000/2,000) for tie point limit. If some images will be not aligned, try 15,000 key point limit.

From GPU utilization perspective it is good to have more key points for matching, because your GPU will be longer time computing and less time transfering data to and from GPU...have more performant GPU(e.g. RTX3080) make sense. If 10,000 key limit would work for you, no need to have  high performance GPU(RTX3060ti would be enough).

You can check my test in this post https://www.agisoft.com/forum/index.php?topic=14622.msg64275#msg64275   to see what is the difference in GPU utilization when different number of key points needs to be matched.

The good is, that estimating location does not take long.

Your detect points 10min. time for 864 photos...you can speed up this phase on CPU with high single core frequency, because it is single threaded task.
I am using RTX 2060 super, intel 11700f@4.4GHz and 18Mpix JPEG files, each ~ 10MB big. My CPU can feed GPU at speed ~ 3-4 JPEGs/s. Time of detect point on one photo by GPU does not care much, because it is quick process, the bottleneck is CPU single core boost frequency.

Try make changes in alignment settings and if it does not help you much, then we will try to change hardware.

Thanks for the help. Reducing the parameters to 10,000 / 4,000 totally helped. The alignment process for the additional 864 photos took 2 hrs 21 minutes for Matching and 52 minutes for Alignment.

Regards,

- brett