Forum

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.


Messages - forumname

Pages: [1] 2
1
Camera Calibration / Re: lens calibration improvement
« on: June 29, 2022, 06:40:28 PM »
Thank you Alexey, I am sending an email. I use 3-5 images of varying quality to understand the lens calibration pipeline.

2
Camera Calibration / lens calibration improvement
« on: June 28, 2022, 05:26:52 PM »
I am trying to calibrate lenses under difficult lighting conditions.

Although metashape finds the chessboard corners, it does not use them (all points grey). Insights as to why the corners are found but rejected?

Using the python API, is it possible to supply the corner locations to the lens calibrator? This would allow corner detection using opencv, or another method like AprilTags.

3
Feature Requests / Re: AprilTag
« on: May 20, 2022, 03:51:09 PM »
Bumping this request as probably one of the most important feature updates. ChArUco would be welcome as well.

4
Feature Requests / request doc.close()
« on: March 11, 2022, 06:45:53 PM »
This has been suggested previously, and would have saved me a lot of trouble because of unexpected behavior with doc.open() and doc.save(). A method for terminating the connection to doc to ensure that all components have been written to that point would be helpful.


5
Feature Requests / Re: The option to only generate depth maps.
« on: March 11, 2022, 06:42:34 PM »
Is this request different from chunk.buildDepthMaps?

see also:
https://www.agisoft.com/forum/index.php?topic=12020.0

6
Can confirm this works on Ubuntu 21.10. Successfully installed numpy and opencv-python.

7
Hi Alexey,

Running from command line does not help. In fact, now the alignment is also not being, although it was before. Intermediate steps like duplicating chunks are still being run. Running the script from the Metashape GUI Run Script... dialogue also does not save the alignment.

I confirmed that the calibration can be manually processed within the Metashape GUI.


EDIT: I found the alignment is a separate issue caused by calling doc.save(path) instead of doc.save(). The depth maps continue to be deleted after creation even if calling metashape from the terminal.

EDIT2: This has apparently been solved. Because of other code, I was calling doc.open(path) before each new block. I guess reopening the document caused a conflict so Metashape didn't recognize the save and treated all the output files as temporary, so they were all deleted as usual when shutting down.

8
Camera Calibration / Re: Calibrate cameras only on markers
« on: March 09, 2022, 05:27:37 AM »
I've been wondering how to do this too. How do you 'delete all points'? I can only find the method for deleting dense clouds.

9
I am attempting to follow the instructions here for Linux.

I navigate to the directory and execute the directed command, but I get an error that the python 3.8 shared objected can't be found

>./metashape-pro/python/bin/python3.8 -m pip install numpy cv2 shutil

./metashape-pro/python/bin/python3.8: error while loading shared libraries: libpython3.8.so.1.0: cannot open shared object file: No such file or directory


I also tried the libssl install as directed, without success.

I am working in a conda virtual environment, but that doesn't seem to make a difference.

10
Python and Java API / depth maps are not saved when looping over frames
« on: February 27, 2022, 08:36:44 PM »
I am building the depth maps through the python API (in Spyder IDE) within a loop, as follows:


for frame in chunk.frames:
    frame.buildDepthMaps(downscale=1, filter_mode=Ms.NoFiltering)

doc.save()


It completes and saves successfully, and says X depth maps were completed. I can verify that the depth_maps folder was created in the project.files directory. However, when I open the project within Metashape GUI, no depth maps are found. Furthermore, if I close the python console, the directories are all deleted immediately. This happens even if doc.save() is called within the loop, or if the loop contains the dense cloud building:


frame.buildDenseCloud(point_confidence=True, keep_depth=True)


However, the depth maps can be properly saved when frames are called manually, e.g.,


chunk.frames[0].buildDepthMaps(downscale=1, filter_mode=Ms.NoFiltering)

doc.save()


The maps are also saved properly if created within the Metashape GUI.

11
General / Re: manually increasing tie points
« on: February 17, 2022, 02:49:24 AM »
I wasn't thinking of tie points as markers, but you're right, that is one option that is straightforward to implement.

12
General / manually increasing tie points
« on: February 16, 2022, 07:10:44 PM »
I am working with a challenging situation with nearly featureless objects and low lighting. Metashape therefore has a great deal of difficulty finding tie points and so depth map generation fails. Is there a way to decrease the threshold for tie point generation to get more initial candidates? Alternatively, can Metashape accept tie points added manually, e.g., using SIFT or some other feature tracker? Thank you for any insight.

13
General / Select face by texture color
« on: August 31, 2021, 04:42:22 AM »
Can faces be selected by color the way points can be? Specifically, after calculating a texture map I would like to remove all faces corresponding to a given color.

14
Feature Requests / identify markers/targets within 4D capture
« on: July 14, 2021, 03:33:02 PM »
Currently, it is very hard to identify the same marker or target across frames of a 4D capture. This is because the markers are not uniquely identified. For example, if I have 500 sequential frames with fiducial markers, 'target 2' will be identified 50 times. If I need to manually place the marker, it is pretty much impossible to tell which of the 50 I should place. If I select the wrong one, I end up with e.g. two 'target 2' on the frame.

The correct behaviour would be to list only the discovered markers (i.e. target 1, target 2, target 3, ...) and to correctly find the marker based on the current frame.

In the attached picture, the problem and the multiple target 2s are shown.

Actually, it's a bit worse because if the point isn't found, then no target/market entry is created for that camera. So it doesn't seem to be possible to add a marker at all.

15
Python and Java API / Re: exportPoints() deletes cloud
« on: June 11, 2021, 02:21:52 AM »
Solved, thank you.

Pages: [1] 2