Topics

1

Camera Calibration / Cannot align more than 2 images at once

« on: October 08, 2024, 06:29:57 PM »

Hi all together,

I have the following problem:
I want to align (just for a test) a very small dataset of 4 images. I have detected the tie-points between the images already, so when I click on "view matches" I can see them.
However, when I select all images and do right click --> Align selected cameras, only two of the images are aligned, but not the rest. Also my log output does seem strange, as the values for the default pair score look strange (everything is 0).

Following things are also important / I noticed:
- I try to match these images with custom tie-points I found myself using lightglue (see https://www.agisoft.com/forum/index.php?topic=16650.0). I'm recreating the p0.ply to pX.ply and the tracks.ply files and add them again to the .files folder. On the first look, the files are identical to the ones Agisoft is creating. Furthermore I can see my own matches in "view matches"
- I know the images have enough overlap, because when I do the complete process within Agisoft Metashape (e.g. using Tools -> "Build tie-points" and then "Align selected cameras") everything is working fine
- The custom tie-points I am using are evenly distributed and also have correct coordinates
- When I just select only 2 images, independent which ones (e.g pair [1,2] or pair [2,3] or pair [3,4] or even pair [1,3] [2,4] I get an alignment.

Does anyone have an idea what is wrong? Or how I can fix this problem? I would love to use my custom tie-points and I found a lot more and for some images I use in my project I cannot find any tie-points with Agisoft Metashape but with my workflow. I've also tried to add markers, but this would only be a Plan B.

Thanks a lot for your help!

Please see here the log output when trying to align all 4 images at the same time:

Code: [Select]

2024-10-08 17:18:40 AlignCameras: cameras = [0, 1, 2, 3], adaptive_fitting = on, use_markers = on
2024-10-08 17:18:40 Estimating camera locations...
2024-10-08 17:18:40 processing matches... done in 0.009425 sec
2024-10-08 17:18:40 selecting camera groups...
2024-10-08 17:18:40 groups: 4
2024-10-08 17:18:40 n groups: 1, total: 4, minmax: [4, 4]
2024-10-08 17:18:40 done in 0.000109 sec
2024-10-08 17:18:40 scheduled 1 alignment groups
2024-10-08 17:18:40 saved camera partition in 0.003687 sec
2024-10-08 17:18:40 loaded camera partition in 1e-05 sec
2024-10-08 17:18:40 processing block: 4 photos
2024-10-08 17:18:40 pair 0 and 2: 812 robust from 815
2024-10-08 17:18:40 pair 1 and 3: 2625 robust from 2625
2024-10-08 17:18:40 pair 0 and 1: 4837 robust from 4838
2024-10-08 17:18:40 pair 2 and 3: 5000 robust from 5000
2024-10-08 17:18:40 pair 1 and 2: 5000 robust from 5000
2024-10-08 17:18:40 evaluating initial pair...
2024-10-08 17:18:40 initial pair evaluated in 0.116213 sec.
2024-10-08 17:18:40 initial pair unstable, considering additional pairs...
2024-10-08 17:18:40 additional pairs considered in 6e-06 sec.
2024-10-08 17:18:40 optimizing initial pair...
2024-10-08 17:18:40 ****************************************************************************************************
2024-10-08 17:18:41 default pair score: n_aligned: 2, n_points_tier: 0, accuracy_tier: 0, reprojection_error: 0, accuracy: 0, n_points: 0, ids: [2, 3]
2024-10-08 17:18:41 optimal pair not found
2024-10-08 17:18:41 initial pair optimized in 0.534593 sec.
2024-10-08 17:18:41 pair 2 and 3: 5000 robust
2024-10-08 17:18:41 adding 5000 points, 0 far, 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 adjusting: xxxxxxxxxx 1.19572 -> 0.921882
2024-10-08 17:18:41 adding 0 points, 0 far, 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 optimized in 0.030927 seconds
2024-10-08 17:18:41 (3 px, 2 3d) sigma filtering...
2024-10-08 17:18:41 adjusting: xxxx 0.921882 -> 0.921862
2024-10-08 17:18:41 point variance: 0.651512 px, threshold: 1.95454 px
2024-10-08 17:18:41 adding 0 points, 65 far (1.95454 px threshold), 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 adjusting: xxxx 0.859036 -> 0.858881
2024-10-08 17:18:41 point variance: 0.606997 px, threshold: 1.82099 px
2024-10-08 17:18:41 adding 0 points, 32 far (1.82099 px threshold), 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 adjusting: xxxxx 0.840306 -> 0.840208
2024-10-08 17:18:41 point variance: 0.593796 px, threshold: 1.78139 px
2024-10-08 17:18:41 adding 0 points, 12 far (1.78139 px threshold), 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 adjusting: xxxx 0.833297 -> 0.833254
2024-10-08 17:18:41 point variance: 0.58888 px, threshold: 1.76664 px
2024-10-08 17:18:41 adding 0 points, 2 far (1.76664 px threshold), 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 adjusting: xxxxx 0.832024 -> 0.832
2024-10-08 17:18:41 point variance: 0.587994 px, threshold: 1.76398 px
2024-10-08 17:18:41 adding 0 points, 1 far (1.76398 px threshold), 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 optimized in 0.250334 seconds
2024-10-08 17:18:41 f 6177.2, cx 0, cy 0, k1 -0.0234523, k2 0, k3 0
2024-10-08 17:18:41 f 6177.2, cx 0, cy 0, k1 -0.0236788, k2 0, k3 0
2024-10-08 17:18:41 finished sfm in 1.17139 seconds
2024-10-08 17:18:41 loaded projections in 8.2e-05 sec
2024-10-08 17:18:41 tracks initialized in 0.001866 sec
2024-10-08 17:18:41 adding 5000 points, 0 far, 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 loaded camera partition in 3.4e-05 sec
2024-10-08 17:18:41 1 blocks: 4
2024-10-08 17:18:41 calculating match counts... done in 0.001604 sec
2024-10-08 17:18:41 overlapping groups selected in 1e-06 sec
2024-10-08 17:18:41 loaded camera partition in 2.8e-05 sec
2024-10-08 17:18:41 1 blocks: 4
2024-10-08 17:18:41 block: 2 sensors, 4 cameras, 18278 points, 36556 projections
2024-10-08 17:18:41 block_sensors: 0.00167847 MB (0.00167847 MB allocated)
2024-10-08 17:18:41 block_cameras: 0.0015564 MB (0.0015564 MB allocated)
2024-10-08 17:18:41 block_points: 0.8367 MB (0.915527 MB allocated)
2024-10-08 17:18:41 block_tracks: 0.069725 MB (0.069725 MB allocated)
2024-10-08 17:18:41 block_obs: 1.6734 MB (1.6734 MB allocated)
2024-10-08 17:18:41 block_ofs: 0.139458 MB (0.139458 MB allocated)
2024-10-08 17:18:41 block_fre: 0 MB (0 MB allocated)
2024-10-08 17:18:41 final block size: 4
2024-10-08 17:18:41 adding 5000 points, 0 far, 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 (3 px, 2 3d) sigma filtering...
2024-10-08 17:18:41 adjusting: xxxxxx 0.974849 -> 0.921862
2024-10-08 17:18:41 point variance: 0.651512 px, threshold: 1.95454 px
2024-10-08 17:18:41 adding 0 points, 65 far (1.95454 px threshold), 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 adjusting: xxxx 0.859036 -> 0.858881
2024-10-08 17:18:41 point variance: 0.606997 px, threshold: 1.82099 px
2024-10-08 17:18:41 adding 0 points, 32 far (1.82099 px threshold), 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 adjusting: xxxxx 0.840306 -> 0.840208
2024-10-08 17:18:41 point variance: 0.593796 px, threshold: 1.78139 px
2024-10-08 17:18:41 adding 0 points, 12 far (1.78139 px threshold), 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 adjusting: xxxx 0.833297 -> 0.833254
2024-10-08 17:18:41 point variance: 0.58888 px, threshold: 1.76664 px
2024-10-08 17:18:41 adding 0 points, 2 far (1.76664 px threshold), 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:41 adjusting: xxxxx 0.832024 -> 0.832
2024-10-08 17:18:42 point variance: 0.587994 px, threshold: 1.76398 px
2024-10-08 17:18:42 adding 0 points, 1 far (1.76398 px threshold), 0 inaccurate, 0 invisible, 0 weak
2024-10-08 17:18:42 optimized in 0.445276 seconds
2024-10-08 17:18:42 coordinates applied in 0 sec
2024-10-08 17:18:42 Finished processing in 1.73741 sec (exit code 1)


2

Python and Java API / questions for adding manual matches

« on: August 15, 2024, 07:50:29 PM »

Hello together,

In my workflow I'm detecting matches between images not only by using Metashape, but also with my own python program. These matches I want to add to a project as well (by using python). As this is not supported yet with a function, I try to apply a work-around by directly editing in the project .files.

So far I've found out the following:
- The matches are saved in the pointcloud.zip, more detailed in the p0.ply to px.ply (every ply-file corresponds to one image)
- Each ply file contains 4 columns: x, y, size and id.
- x and y describe the coordinates of a point and id is a unique identifier (integer) for a position/match. Points from different ply files with the same id point to the same position/are matches
- Furthermore there is a tracks.ply with just one column "color". The number of entries are similar to the number of matches. However, the same number can be multiple times in that ply file, so the entries in this file are not unique.

I would have two questions now:
- What does the column size in the pX.ply file stand for? Is it a quality metric?
- For what is tracks.ply required? It is somehow also required for matching, because if I fill it with default values (e.g. just with the value of "1"), I see my own matches in Metashape, but the alignment then fails.

Thanks a lot for help!

3

Bug Reports / Ubuntu GUI not working

« on: June 05, 2024, 03:05:42 PM »

Hello together,

I've downloaded Agisoft Metashape 2.1.1 Professional for Ubuntu from the website. Before starting I've installed libxcb-xinerama0 without any problems.

The program is starting (by executing the metashape.sh), but unfortunately something is not working correctly for the GUI: Elements (like input fields) are only partially shown or parts of the GUI are not rendered correctly (they are hidden by a gray box and only popping up when I hover over it, but when I hover away, they disappear again. I've included a screenshot to show the problem.

Any ideas what I could do to fix that issue? 

Thanks a lot for your help!