Messages

1

General / API + Documentation Issue: CaptureDistance

« on: July 25, 2025, 01:47:06 AM »

chunk.capture_distance

 
Alexey-
The current Python API docs for Metashape 2.2.1+ still include the property:

Code: [Select]

chunk.capture_distance however this does not exist for the past few versions.

I've also attempted to set this at the chunk-level using

Code: [Select]

chunk.meta['CaptureDistance']` Unfortunately this doesn't have any effect on the chunk's Capture Distance.

Some time ago you added an additional option of setting the capture distance of each photo.  I don't believe that's in the docs yet.  Regardless, I'm hoping to understand how to set this at the chunk level.

This is a two part request:
1. Can you explain how `chunk.capture_distance` is to be set in Metashape Python API for 2.2.1+
2. You may want to update the documentation, removing the old approach and replacing with the answer from this post?

Thank you in advance

2

Python and Java API / API Use: alignCameras > min_image

« on: July 06, 2025, 03:50:15 AM »

Back in 2019, Metashape released 1.6.0 and added the attribute `min_image` to the `AlignCameras` function.

I realize this sets the minimum projections per tie-point to be created during the Alignment, but I don't believe this exists anywhere in the UI to test without using the API, is that correct?

I've often wondered if using this is valuable in high-overlap jobs.  My thinking is that if I set the minimum higher, then the tie-point limit is only used by higher quality tie-points, whereas without the limit (implicitly indicating a minimum of 2 photos per tie-point, minimum) -- the tie-point limit could ?theoretically? cutoff higher quality tie-points while keeping lower quality tie-points.

However, there is a chance that Metashape locates tie-points in order of those with the largest number of matching photos first, in which case, this setting would not have much effect on the quality of photo alignment.

In that event, I wonder if it's better to ignore this parameter and only reduce tie-points by photo-count after the alignment is complete.

@Alexey - can you share any more insight about this parameter?  For example, in high-overlap jobs with a tie-point limit in the alignment stage, would it be useful to fill the limited tie-points per image with higher quality points?  Or would it be better to ignore this parameter any only use it once the photos have been aligned, in the gradual-selection stage with iterative optimizations?

Thank you in advance

3

Bug Reports / Re: merge chunks + masks fails

« on: April 23, 2025, 12:42:03 AM »

Thank you Alexey, that's a huge help.

Have you written anything about this prior, anything that goes into more detail about the changes?

Also, will the API calls be changed as well / will I need to update my gradual selection scripts in Python?

Thanks again!

4

Bug Reports / Re: merge chunks + masks fails

« on: April 22, 2025, 10:12:10 AM »

I just attempted to install this pre-release and the Model > Gradual Selection appears to be missing?

I have confirmed that I'm in the sparse point cloud.  To clarify, it's not merely greyed-out, the option is completely gone?

5

General / Re: Control points with aerial surveys

« on: December 27, 2022, 01:21:34 AM »

@Apavlicek - I'm going to answer based on my understanding of your question.  I'm concerned I may be misunderstanding your intention and if I've answered wrong, please let me know and I'll clarify.

In Metashape, all Points (Lat/Lon coordinates) used for reference are stored together.  You can choose to make them either "Control Points" or "Check Points"

When you choose to make them "Control Points", they are used to rectify/correct the alignment to achieve better accuracy.

When you choose to make some of these "Check Points", these will not be used to rectify/correct Metashape's processing.

Imagine you have a project shaped like a vertical rectangle.  Imagine you have 6 reference points in that image (four near the corners, one near the top-center and one near the bottom center).  In this example, imagine you set your corner points as Control Points.  You then set your center top/bottom points as Check Points.

The four Control Points will help rectify the output and if everything is correct, the two Check Points should be matched.

If they are not matched, there are two things you can do with Check Points.  One, you can use them to help determine where there is an error elsewhere, keeping them as Check Points.  Two, if you are 100% certain the Check Points are more accurate than Metashape's output, then you could choose to convert them into Control Points (forcing the output to use them to correct its output).

If the Check Points match, there is nothing more to be done with them.  They are not used, or needed, and nothing else you will do in the future with those points (other than re-confirm if you move anything around and re-align in the future).

Does that make sense?

6

General / Re: Agisoft Metashape 2.0.0 pre-release

« on: December 27, 2022, 01:10:17 AM »

I've noticed something strange in v2.0 beta (I've updated to the latest Dec 25th release and the problem remains).

In both 1.8.5 and 2.0 I have enabled the option:
Preferences > Advanced > Load camera orientation angles from XMP metadata

In 1.8.5 this loads Yaw, Pitch, Roll from the XMP metadata.

In 2.0 this only loads Yaw.  Pitch and Roll are always 0 (zero).

It's strange that it does load Pitch, but not the other two.

Thank you.

7

General / Re: Best Handheld Camera for Facade photogrammetry

« on: October 16, 2022, 06:29:46 AM »

Andrew,
Something with a large sensor, either full-frame or APS-C (cost-effective/common example: a5100) with least lens distortion (the largest POV where you get the least distortion, such as a 40mm on a full frame sensor).

Good luck!

8

General / Re: Agisoft Metashape 2.0.0 pre-release

« on: October 16, 2022, 06:27:08 AM »

Pre-release version of Agisoft Metashape 2.0.0 is available for download.

Dmitry,
Thank you for the hard-work and new release.

Normally there is something relatively significant in the differences between full version increments (1.xx vs 2.xx).

Is this this case with the new 2.xx version, or is this just a natural progression, much like 1.8 to 1.9, for example?

From the perspective of the changelog ( https://www.agisoft.com/pdf/metashape_changelog.pdf ) - it appears like this is a standard sub-version upgrade and there are no major paradigm-shifts going to the new v2 version?

Thank you in advance!

9

General / Re: Nvidia Tesla K80. A good, cheap option?

« on: October 01, 2021, 10:13:21 AM »

There was another post (some time ago) inquiring about another NVIDIA Tesla card, possibly the K40.  The consensus was that despite the age, it does benchmark similar to modern cards, despite the high power consumption.

Keep in mind; however, that due to the power-consumption, it's easy to heat-up which results in *throttling* down the GPU.  Therefore, in consideration for the high power consumption, you must be sure you add supplemental cooling.  Either on the card itself or adding some high-power on-demand ancillary cooling.

Keep in mind the K80 also has 2x 12GB RAM (it's essentially designed as a 2x GPU on a card).  Because GPU memory is so crucial to Metashape, you should notice that for the price, you'll get great bang-for-your-buck with the K80, providing you don't mind compensating for the power/heat issue.

If you do go with a K80, will you report back and share your real-world experience?  Thank you!

10

General / Re: How to run a .py Script

« on: September 30, 2021, 09:07:02 PM »

I often Batch process these things for multiple scans overnight.

Hi Brian, I couldn't agree more.  I do the exact same thing (and I assume many others do as well).

Because "Gradual Selection" is so integral in the Metashape workflow, I am surprised it is not part of the "Batch" features.

My assumption has always been that it might not be, due to many people changing the values based on the output of each.  No two projects are the same and to get the best output, people often tweak the amount the "gradually select".

None the less, there are many times where we might be re-running a project where we know the values, or running an initial rendering/test to get masks or an idea of where things lie, etc.

For all these reasons and more, I do agree, it's still very valuable to have gradual-selection added to the batch tool!

11

General / Re: Get more accuracy by using Survey Points

« on: September 30, 2021, 09:03:34 PM »

Select an area in the Point Cloud (Tie Points, Sparse Cloud -- not the Dense Point Cloud).

Choose to show photos for this area (Right Click on Selection > Filterr Cameras by Tie Points)

On the resulting images, choose a very reliable location to place your "marker" (Metashape's term -aka- "survey points" in your vernacular).

Choose an area that is blunt and preferably true horizontal or true vertical (or close to it).  I've had issues on smaller areas and less explicit elevations (eg small slopes, such as an architectural detail on a structure).

Right click on the photo and choose "Add Marker" to get a marker with coordinates.  Choose "Place Marker > New Marker" to get a point without coordinates.

NOTE:  Maybe Alexey can confirm - I am not certain why the "New Marker" is void of coordinates or how this is intended to integrate in Metashape's workflow.

You can change the coordinates if you like.

If you DE-select that "marker" in the list, it will not be used for purposes of rectifying/correcting your geo-coordinates -- but it will be used to help tie photos together.  If you keep it selected, then it will be used to correct coordinates, which may not be what you want.

Page-down through the remainder of your photos (the photos that are restricted to that area / tie-points you had selected).  Make sure Metashape has placed the marker in the same location on all subsequent photos.  If it has not, then you should move the marker to be in the correct location.  In addition, in the sections that have the most clear-view, set those manually for at least 2 more images, even if they are not wrong.  This just helps Metashape confirm the location with greater assurance.

I hope I understood your question properly.  If I have not, can you let me know how I've misunderstood and I will attempt to clarify or change my answer based on your unique needs.

12

General / Re: How to run a .py Script

« on: September 30, 2021, 08:48:54 PM »

Hi Brian,
Although I don't have a suggestion to get around the limits of the "Standard" version -- the steps taken by the script you linked are not exceptionally time-consuming.

The gradual selections are obvious near real-time and although the first optimization takes a couple minutes for a couple thousand photos, subsequent runs are very fast.

One key thing to consider / speed the process:  When optimizing -- the script does not check the boxes for B1, B2, K4, P3, P4, Adaptive, Covariance, or Additional Corrections (for the first two gradual-selections).  On the third gradual-selection (Reprojection Error) you will select the Adaptive option (disabling all others).  With these settings, your first two attempts should be expedited and your third attempt will be faster, despite the extra processing, because the images are much more optimized prior to the operation.

In short, while I agree, it would be extremely valuable to add "Gradual Selection" to the "Batch" feature, at least you can replicate this script without too much time.

Also, this script appears to use Reconstruction Uncertainty at 15, Projection Accuracy at 15 (typically I believe 5 is a better range for this, if it doesn't remove too many tie-points), and Reprojection Error iterates up to four times to reach 0.20 starting at 10.

Sorry I don't have a better option, but I hope this helps