Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Topics - Mariusz_M

Pages: [1] 2
I have been placing markers on a model that is made of around 8000 photos, which is not that big compared to other models I work with. As long as I place a new or existing marker on a photo (when everything is already aligned), it works well. But when I accidently press "add marker" instead of place marker, it goes through every single photo and it takes ages. I would simply like to abort adding a marker when it takes a long time by pressing "Esc".

I am working o a model made of 8000 photos. The model has 50M triangles.

 Now each time I want to process 10 textures it takes ages, because first it wants to do it on CPU instead of GPU. Even if I click processing in the background and it closes the model, for some reason it does not have enough momory to do it anyway. The part that takes ages is blending textures, so after UVs are done and I do not udnerstand why the size of the model matters at this stage. Creating 10 textures on a 3-5M triangles takes only a few minutes.

A huge part of the process is "Estimating Quality". Is is the same as the option of Estimating Image Quality? I ran Estimate Image Quality on all images just before texture generation and it took only around 35 minutes. Did not help though. Now at the stage of blending textures it is estimating quality again. It has run for the last 6 hours and so fat 72% done.

Feature Requests / Second layer of masks
« on: May 03, 2024, 02:06:30 AM »
In underwater photogrammetry quite often I need to mask either divers or bigger fish. The masks must stay on during the model creation, so no diver or big fish affect depth maps. Because of that, I am unable to use the volumetric masks at the same time, so most of the time. Volumetric masks are quire useful to fix some small problems of shipwreck parts.

The solution is quite simple. Please introduce a second layer of masks. Then I will be able to draw separate m asks for actual masking, and another independend layer of masks for volumetric maks. Then during the model creation I would be able to select: Depth maps - layer 1 masks or both, Volumetric masks - Layer 2 masks or both. Quite simple solution and I guess from the programming side also not very difficult.


There are several cases when prioritizing closest photos for texture generation would be prefered. There are also cases, when it would be prefered to select a specific camera group and prioritize them in texture generation. So please implement these options if possible. Below only two scenarios.

Case 1.

Flying a drone around a building. First flying far away, to get the overview and to get informaion what is around the building. Then flying much closer only to focus on the details of the walls. At the moment I have no control which photos will be used for generation of textures on the walls and in some cases the ones from far away will be chosen, although there are closer photos that better show the details I care about.

Case 2

I have underwater photosets where I scanned a wreck. Most of the photos are taken with a 20mpix GoPro and they help building the model well. But there are also some photos from a higher resolution and quality DLSR camera showing only details up close in those parts of the wreck viewers will most likely zoom in and expect details. At the moment I have no control which photos will be used for textures. In this case I would prefer to select a camera group from the DLSR and set "prioritize in texture generation".

Recently a new option was added to split models in blocks. There is not much about it in the manual. What would be the use of it and how much does it differ from building a tiled model?


I have a photoset from underwater without GPS co-ordinates recorded with an autonomous boat. The same boat has a sonar with GPS and saves a track. I would like to use the CSV file from the sonar which contains the track (trajectory) of the boat to preselect and reference the photos. At the moment it does not seem possible, as the "import trajectory" function is for laser scaners.

So Metashape can already load the boat trajectory and displays points with GPS co-ordinates and time. Now all that is needed is Metashape to also read the timestamp on each photo during the alignment process and place it on the same trajectory based on the time the photo was taken. There could be another option to the existing Reference Preselection, like Reference Preselection: Timestamp

This simple addition can help with underwater photogrammetry a lot. Since nowadays more and more underwater navigation devices can output a track which sometimes has lower accuracy than similar devices above water, but could easily be used for preselection of big datasets and rough reference of the whole model

Feature Requests / Pause/play button for background processing
« on: December 15, 2022, 04:23:07 PM »
I often process big projects and they run in the background for a few days. Every time I want to pause processing and use my computer for something else, I first need to re-open the project to pause it. It would be so much easier if I caould just pause/resume background processing with one click without loading the project abck into the memory.


It has actually been happening for the last few versions and sometimes is quite annoying. I have an integrated Intel UHD and Nvidia RTX 3070. So obviously I only want to use the Nvidia adapter and this is the only one switched on in the preferences. Hovewer, from time to time, quite randomly, I see in the console that Metashape uses both GPUs. Then I go to preferences and see both GPUs ticked. So I untick one and work on the project for several days and then it happens again.

Also, even when Intel GPU is switched on, it still seems to be used during depth maps generation alog with the NVIDIA GPU. Is it normal?

Feature Requests / Add UNDO support for accidently deselected selections
« on: February 01, 2022, 09:22:39 PM »
Sometimes I use the selection tools on a mesh, creating a quite complicated selection using SHIFT and CTRL keys. But if I accidently do not press either of the keys, all the selection gets lost and I need to start again. UNDO does not work in this case. Can you please change it? So next time I accidently deselect what I have spent 20 minutes on selecting, I will just undo deselection and continue selecting.

I use intelligent scissors a lot and I have a high DPI screen. So the points created with the tool are very small. It does not affect the process so much, but when I want to close the poligon, I actually need to zoom in a few times and it is still difficult to click exactly where the starting point is. If the points were bigger or it was possiible to click near the point to auto close, it would be so much easier.

Also, the auto close option should be available in the right click menu. So every time I wanted to close the selection created with the tool, I would just use the option auto-close and Metashape would just connect the last point with the first a straight line, even if they were far apart. So much easier.

I use the magic wand tool for maskig a lot. And since the beginning, to change the tolerance of the tool, I have to go to a separate tool/window, set the new value and come back to using maric wand. Can you please make the tolerance setting available every time the wand is used, so I can easily readjust the tolerance each time I need it?

Quite often, when processing textures, I end up with some low quality patches. It happens only because Metashape chooses a more distant photograph to use in that particular spot. Could you pleas add a tickbox to texture generation box that would prioritise the closest and therefore most detailed photos. It would save a lot of manual work I need to do with disabling the farthest photos at texture generating stage.

Bug Reports / Standard 1.8 crashes windows during mesh generation
« on: January 19, 2022, 11:26:57 PM »

I tried to proces a mesh(from depth maps) under Standard 1.8 in a project that I had processed multiple times under earlier versions. The project has 3652 photos aligned. It crashed windows twice exactly at the same stage of the process.

2022-01-19 19:30:55 Marching cubes: part #3/7...
2022-01-19 19:31:11 Data preloaded in 16.411 s (1381 MB -> 4738 MB, 288.766 MB/s)
2022-01-19 19:31:55 cubes part marched in 32 mini parts: 1.56597e+06 avg cubes -> 547732 avg faces -> 158021 avg decimated faces = avg ratio x3.46619, done in 60.381 s (28% cubes IO + 2% sorting + 50% marching + 19% decimating + 0% saving)
2022-01-19 19:31:55 Marching cubes: part #4/7...
2022-01-19 19:32:14 Data preloaded in 18.668 s (1663 MB -> 5703 MB, 305.517 MB/s)

The full log from that session is attached as a ZIP file, because it is over 5MB. The log from previous session is exactly the same. Crash happens during part #4/7.

There is no trace of a problem in Windows event log and at both times I was not looking at the screen when it happened. I just came back after an hour and my computer has already rebooted.

So I unistalled 1.8 and re-installed 1.7.5 and run the same process. The problem did not occur.

I have Nvidia 3070 and 32GB of RAM. I have successfully used this setup for projects that have 45000+ photos under 1.7.5 and earlier vesrions. So it is not the matter of my computer or the number of photos. It is just some issue with version 1.8. I cannot be more precise, because there is no trace what went wrong.

General / Processing a massive underwater photoset without GPS
« on: October 23, 2021, 09:57:39 AM »

I am processing a massive underwater photoset (so far 50 000 photos) and I keep taking 10 000 photos every day. There is no GPS data and I am looking for a solution to speed up the matching part of the process. Therefore, I need to understand how each of the following work.

1. Reference preselection set to sequential + generic preselection. What is it going to do? Will the program do the sequential preselection first and only the photos that have no pair will have generic preselection applied? Or will it do the generic preselection for all photos, finds all matches and then also add sequential preselection on top to find even more matches and actually extend the length of the process?

2. Reference preselection set to seqential while every second photo disabled, so only half of the photos are being aligned. Provided that they will align somehow... Then I enable the other half of the photos and what are my options then? Sequential again? Or perhaps Estimated? With generic preselection on or off?

Any other ideas how to match only the photos that should be matched?

Bug Reports / 1.7.1 Mesh colour disappears when closing holes
« on: January 29, 2021, 12:07:11 AM »
Now when I now close holes on a coloured mesh, the colours disapear and I have to reprocess them again. It used to close the hole and the new faces had colours from neighbourhood triangles. But now all colours are gone and I need to colorize again.

Pages: [1] 2