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.


Topics - FIFER

Pages: [1]
1
General / pount cloud presents points in a double-layered arrangement
« on: October 12, 2023, 09:06:49 PM »
Hello,

Working on a dataset of downlooking photos over the seafloor.
The dataset was acquired with the lawn-mower method and the images well overlap between eachothers.
I am also quite sure there are redundant photos but I did not try to discard them.

The alignment gives me a quite good result but I notice a weird effect for the first time with this dataset: in many occasions it seems like if the point cloud consist of two layers of points reproducing the same surface. The 2 layers are slightly separated and one over the other. In some sense it is like if one would suffer from a "doubled-vision" problem.

Thinking that this could be an issue of having too much overlap between adjacent images, I did try to reduce this redundancy by "Reset Alignment" on a set of adjacent cameras but this double-layer effect still persist.

Any idea of what is causing this and how to correct it?
Would it be something that the recursive optimization will fix (still have not tried)?

Attached some of screenshots to show the double layers as well as the track of camera positions (blue squares size shrinked to make the underlying cloud visible).


2
General / GeForce v RTX (ex Quadro)
« on: June 12, 2023, 11:55:07 PM »
Hi,
My use of metashape is mainly to create Orthophoto and DEM, which will represent primarily the data I will work with later on. I might still want to create texture on 3D models for show, but it is not the main aim.
 
So far, my projects range between 3k-30k images of 20-27MP per project (I know… too many… you use less… but I need that for overlap).
I am looking for advice on upgrading my GPU (currently a Quadro P2000) to speed up the processing.

My current work station has the following specs
CPU : 2x Intel Xeon Platinum 8160 (total of 96 threads)
Ram : 128 GB DDR4 ECC (8x 16GB 2666MHz 2RX8)

At the moment I have been proposed these 2 alternative solutions:

1.   NVIDIA RTX A4000 (16 Gb GDDR6 VRAM; 6,144Cuda cores)

2.   NVIDIA RTX A4500 (20 Gb GDDR6 VRAM; 7,168 Cuda cores)

However, reading in this answer on the forum (https://www.agisoft.com/forum/index.php?topic=14202.0) @Alexey was suggesting that GeForce RTX 3080 or 3080 Ti rather than a RTX A5000 because it “should show better Metashape performance” if “not planning to generate extra-large textures”.

Now, I see that currently a NVIDIA GeForce RTX 4090 has more CUDA cores (16384) and more memory (24 GB of GDDR6X VRAM) than the 2 I have been proposed, for apparently the same price of the A4000.

Is there an advantage in choosing a GeForce over a RTX (ex Quadro)?
What will be the advantage of the RTX over the GeForce?

I am not a gamer so take that out of the equation.

Thanks
Filippo

3
General / update from 1.x to 2.x Professional edition
« on: April 26, 2023, 11:05:21 PM »
Hello,
I want to update from 1.x to 2.x Professional edition.
From what is written on the web site:
- it should be free
- I need to install the 2.x version in the same folder of the 1.8
- I need to reactivate the license

I did all these things but when I enter my license key it says it is not valid.
Do I have to deactivate the 1.8 before updating to the 2.0?
What can be the issue?  I read that there is a new procedure but it is not clear to me what that is.

Thank you.
Filippo

4
General / Error: clEnqueueNDRangeKernel
« on: January 04, 2023, 06:32:30 PM »
Hello,
I was running the align Photo step on a large dataset  (27k pics).
I got this error some 12 h into the pair selection step.

Error: clEnqueueNDRangeKernel(queue(), kernel.kernel(), work_dim, global_work_offset, global_work_size, local_work_size, 0, NULL, &ev): CL_INVALID_GLOBAL_WORK_SIZE

What could be the cause?

Attached the log of the processing

5
Hello,

I have a big dataset of images (>26k) taken with a lawn mower pattern and roughly facing down.

My aim is to speed up the processing time.

I am experimenting on a small set of this dataset to figure out an approach that will considerably speed up the processing time. I am testing 3 approaches:

1) Attach estimated GPS (the dataset is originally non-referenced) position to cameras and use Preselection source in just 1 chunk.
2) Split the non-referenced dataset in different partially overlapping chunks, Align within chunks, then Align and Merge chunks
3) Combine 1 and 2: divide set in chunks and use GPS references within chink to speed up.

Please read the attached pdf which details the output of my tests and provide well described screenshots of point clouds (sorry but would not know how to do otherwise).

As a teaser, here are my

Questions:

1) Is using the Preselection Source with GPS position the only way to speed up Processing?
2) Should the use of chunks help speed up the processing? Or
3) Is the use of chunks meant only to address memory problems but it is inherently longer than single chunk processing?
4) Why point based chunk alignment do not reuse already detected key point and need to redo this step?
5) When aligning chunks, Should a “worst” alignment than that obtained in a single chunk be an expected typical result?
6) Am I using Chunks in the wrong way?



Thank you very much for any explanation.
Cheers
Filippo

6
General / Align with GPS coordinates: should images be checked or not?
« on: December 08, 2022, 06:13:49 PM »
Hello,

I have a big dataset of images (>26k) taken with a lawn mower pattern and roughly facing down.

I have estimated the GPS position for the majority of them and imported it in the hope to speed up the alignment.

To do so I understand I need to set the Reference preselection to "Source". However it is not specified if the cameras with the GPS coordinates need to be "checkec" or "unchecked".
I have been taught to leave images, markers and scale bars unchecked images  before Gradual Selection for Reprojection Error, and thus not during alignment.

So, I am wondering if the position of unchecked images is still used to speed up the matching and image pairing during alignment with preselection set to Source.

Otherwise, Should the cameras be checked during alignment and then unchecked (to avoid constrainig) during the first steps of the gradual selections?

Thank you
Filippo


7
General / AMD GPU not used during mesh building on VM
« on: November 07, 2022, 09:39:34 PM »
Hello,

I am working on a project for which i obtained a sparse cloud of some 900K points obtained from some 9000 pics.

I am using a Windows Virtual Machine on Azure with
- AMD EPYC &V12 64-Core Processor 2.44 GHz (although I think it says 32 virtual processors in the task managers)
- 112 GB RAM
- GPU AMD Radeon Instinct MI25 MxGPU (with  16 GB dedicated GPU Memory and 72 GB of GPU Memory, according to Task manager infos), with DirectX 12 (FL12.1)

I recently tried to build a Mesh using the depth maps and I monitored the use of the GPU in the task manager on my VM.

I have the feeling the that metashape is not really taking advantage of this GPU: a part from the fact that it took a lot of time to process, I almost never saw the GPU used more than at most 9%. However from the details provided by Metashape in the processing window I could see that the GPU was found and "invoked".
 
Also, One thing I noticed as a message is that at some point metashape look for something called SPIR for the GPU and it doesn't find it.

I took some screenshots of several stages of the Build mesh processing and I am providing them here https://drive.google.com/drive/folders/1DMjuDqHigAxNyPITI93Aq1yQa_Lb-xmQ?usp=share_link, along with the log file of the process.

Before you ask: I did not checked "use CPU" in the GPU settings but just selected the GPU.

Additionally I found this performance test on line (https://www.pugetsystems.com/labs/articles/Agisoft-Metashape-1-6-5---AMD-Radeon-RX-6800-XT-Performance-1991/) where 2 AMD GPUs were compared to 2 NVIDIA GPUs. As you can see from the table and graphs in there, It seems that the time required by the Build depth maps step is almost not different when using AMD GPU or no GPU at all...

So my question is :
- is my VM taking fully using the GPU when doing the Mesh depth maps in my project or not?
Put it in another way: My VM equipped with a AMD Radeon Instinct MI25 MxGPU a good machine or should I really insist to have a NVIDIA card?

8
General / AMD Radeon GPU on Virtual Machine
« on: October 19, 2022, 04:43:09 PM »
Hello,
I am starting to use Metashape 1.8.4 on a Virtual machine on Azure. The VM has Windows and it is equipped with a AMD Radeon instinct MI25 mxgpu.
Can you confirm that this GPU is compatible with metashape?

I am confused because the Agisoft Support page on compatible GPU (https://agisoft.freshdesk.com/support/solutions/articles/31000150614-general-information-related-to-gpu-processing) does not list this AMD gpu, whilst it list the NVIDIA tesla M60 which was another option for the VM (I couldn't choose the GPU for the VM).

However, I saw another user post referring to the use of this AMD gpu.

Is there any reason to prefer NVIDIA rather than AMD on VM?

the Gpu drivers have been installed on the VM. Is there anything to do in Metashape (except select the GPU in the preferences) for it to use the GPU?

Thanks

Filippo

9
General / using more than one pcu
« on: June 27, 2021, 05:11:30 PM »
Hello,

I have a professional node-locked license, and I have the chance to use a desktop computer that has more than 1 cpu.
Is it possible to have metashape use all the cpus available (e.g. parallel computing)?

If yes how?

10
General / Georeferencing after creating orthomosaic
« on: March 05, 2021, 07:24:38 PM »
Hello,

Background info: I'm mosaicing underwater transects, so I do not have images with GPS coord attached. What I usually did in the past is to create a mosaic just from the pics, export it, import it in QGIS and georeference the mosaic there using the coordinate of one end of the transect and the estimated coord of the other end. CRS use was EPGS 32619.

Current situation: Now I'm in the same situation as background but also have few markers in the transect of which a I know the position relative to the transect.
- In metashape I built the mosaic as before but this time I also placed the markers on the pics before alignement with relative coordinates in meters (e.g.,  start (0,0); end (0,30); pt1 (2,10); pt3 (-2,15);...).
- used patch and Assign picture to fix mosaic' part that may be blurried
- I export the mosaic and import it in QGIS for referencing.

Problem:
In QGIS for some reason I can't georeference all the mosaics I have. So I was looking for a way to georeference them directly from metashape.
Is there a way to assign the real world coordinates to markers and "update" the mosaic (without rebulding it) so that it can be exported in the CRS I need? I'd love not to have to rebuild the mosaic since I have spent quite some time on some of them improving their aspect in many places (i.e., procedure patch-> Assign picture-> mosaic update).

Thanks


11
Bug Reports / TIFFWritescanline error on mosaic export
« on: February 08, 2021, 06:10:05 PM »
Hi,

After doing some correction to a mosaic (i.e., manually assign pics to some areas using the patch tool) I tried to export the mosaic with the JPG compression.
At this point the export fails around 4% and I get an error that says "TIFFWritescanline error".  However I was able to export the mosaic with the LZW compression (which produced a some 2Gb file).

Any idea of what is going on and how to be able to use the JPG compression again?

Thanks
Filippo

Pages: [1]