Messages

31

General / Re: AMD 7900XT low gpu usage on Align Photos and Build point cloud

« on: October 16, 2024, 10:47:03 PM »

@tutos You need better CPU to feed all 3 GPUs. Also photo resolution matters. AMD 7900XT is probably in FAN less mode under 60°C.
One 7900XT(~51 TFLOP/s) should be more performant in build depth maps than 2x 2080s(2x ~11 TFLOP/s). Try to run only 7900XT  and check the times. If it beats two 2080s you can sell those cards and save tons of watts. Or buy better CPU.

32

General / Re: Laptop for photogrammetry

« on: September 11, 2024, 08:58:13 PM »

All 3 laptops will behave the same in terms of performance, because the limit on laptops is only the temperature(CPU/GPU).
If you undervolt the CPU/GPU on any of these laptops, then you will see much better performance than on the rest two laptops which will be only on default.
I would personally chose the 7940HS, because it is manufactured on 4nm TSMC process... best performance per watt. Also RTX4060 is 4nm TSMC process.
If there are no plans to undervolt CPU/GPU, then buy that laptop which has best cooling(after watching some youtube disassembling videos).

33

General / Re: Photo Alignment produces very strang output result

« on: September 03, 2024, 04:26:01 PM »

For this you will need camera with a lot higher resolution of sensor than just 1200x1200.
With better camera 18/24Mpix Metashape will be able to recognize also feature points from white plaster walls, if they are not completely flat.
I would suggest you to change the path/positions from where you were taking the photos, because you took a lot of pictures which are useless.
To shoot the white walls you should be ~ 2-3m far away from them(in the middle of the benches) and point the camera perpendicular to wall and also take photos from oblique angles. You took the photos from too close and too far distance only.
You can try put some markers on the walls or at least to have some bench part on each photo to keep the continuity of shots.
It is interesting you were able to get almost correct result in Copy of Chunk 2.

So try to focus more on the surface you really need, choose the correct distance from the objects/surfaces and that will set you on the right path(positions from which you will be taking photos).
With better camera choose only High alignment quality, which is 1:1 to original resolution and set some tie point limit(2000-4000) to speed up the process.

34

General / Re: Is smoothing a point cloud causing this?

« on: August 21, 2024, 02:25:35 AM »

I did exactly the same.
What I noticed from your video, the number of points in your case has increased from 1,819,558 to 1,820,626 after smoothing, in my case the count remains the same.
Also your coordinate precision is 0.1mm, my 2.49mm.
So I think there are some additional data in your point cloud(Point attributes) which smoothing takes into account. I have only Color information 3 bands, uint8.

35

General / Re: Is smoothing a point cloud causing this?

« on: August 20, 2024, 09:36:32 AM »

I downloaded ply file from sketchfab and there is no problem at all.
I tried values 0.1m, 0.2m, 0.3m and 0.5m. Larger values does not make sense and everything works as expected, noise in point cloud disappeared after smooth.

36

General / Re: Is smoothing a point cloud causing this?

« on: August 18, 2024, 11:09:44 PM »

Maybe there is a problem with normals, can you share few meters cropped area with us?

37

General / Re: Is smoothing a point cloud causing this?

« on: August 18, 2024, 09:25:26 PM »

Did you try different values for smoothing?
Does your original point cloud have ~evenly spaced points, so they are not in some kind of ~grid structure?

38

General / Re: Metashape node does not use 2 GPU even with gpu_mask 3

« on: August 11, 2024, 10:16:35 PM »

Try to spawn another worker node on that machine.

39

General / Re: Maximise my iMacs processors for MetaShape

« on: August 08, 2024, 09:54:22 PM »

You are using only CPU for detecting points and that process is not using properly hyper threaded cores(so only 8cores/8threads and not additional 8 hyper threads).
Your speed is ~4.6 photos/s. Try to enable GPU for processing in preferences GPU tab and it will use GPU for detecting points ...it should be faster - but photos needs to be decoded and that is single threaded operation, so do not expect miracle.

The only way how to masivelly speed up this process is use Metashape PRO and run several workers in local network processing on the same computer... then photos will be decoded using all your CPU threads and GPU detecting points will be several times faster.

40

General / Re: Correct way to maximise processing speed?

« on: August 05, 2024, 03:43:11 PM »

@jrp
I think you do not need to use linux boxes. Just use your normal host machine and run one server node and several worker nodes according Metashape manual(page 165).
For controling number of active workers you can use Monitor app or also server WEB interface(manual page 167).
Spawning multiple workers can be done using linux shell script.
How to automatically  change number of active workers is tricky. You can try to check periodically WEB interface log using javascript(detect when one subtask is ending and another starting) and then manipulate web HTML components using javascript. Maybe there are some commandline options for monitor app or Metashape network processing or python syntax.

41

General / Re: Correct way to maximise processing speed?

« on: July 24, 2024, 01:30:25 AM »

@geometrical
Yes, exactly, spawn many workers on your pc and then manually adjust the active number according subtask. For each subtask the optimal number is different.

42

General / Re: Agisoft Metashape Pro on Server wit NVIDIA L40s

« on: June 27, 2024, 08:41:49 AM »

Yes, the license is bound to computer on which you can run as many local instances as needed. Even the standard version can be run several times at the same time on one PC.

Add just more workers on one PC is the key to speed up single threaded tasks(detecting points) and cover the gaps when GPU is feeded with data or is preparing the shader for run or is allocating VRAM and is not processing at that time. What will be needed is manage(stop and resume them according needs or according actual VRAM usage) local workers in Metashapes Monitor application.

I would still check on your existing HW how it is ballaned for your specific project(metashape settings, photo resolution, allocating VRAM...) and then it would be easier to guess how many cores CPU  should have on new server and if buy one L40 or two, or some RTX GPU if 24GB VRAM would be enough.

43

General / Re: Ultimate PC Metashape Processing 20k+ images 2024?

« on: June 26, 2024, 05:25:01 PM »

No need to go with just medium quality in alignment settings, always use high(if photo quality is good). It is good to take smaller amount of photos from your large project and test how low parameters(key/tie point limit) still work for you. Check Metashape PDF manual what is saying about each parameter in options.

Quality of depth maps depends on your photo resolution, distance from object and how much precision you need in final result. Ultra high gives you maximum from your photos. On High you are using just half of the photo resolution. On Medium just 1/4.

i9 is just name of CPU series, what really matters is the number of cores(performance + efficient) and the resulted frequency on which they will be running at full utilization. So new CPU/GPU/RAM will finnish your big project faster, but not as much as you are expecting(from few weeks to just few days). If you want to see massive time savings, you need to be 100% sure that your parameters numbers are right chosen(not unnecessarily high and still sufficient for expected final result) and then it makes a lot of sense invest into new HW components hundreds or thousands $€.

Also you can let us know more about your project, what kind of photogrammetry it is, how big the photos are, what you need to produce as a final result(point cloud, mesh with textures, orthophoto maps., ....)

44

General / Re: Depth Maps Turning Black with Large Drone Dataset

« on: June 26, 2024, 11:23:14 AM »

Strange, did you try also only High alignment quality, which is 1:1 original photo resolution? I am not sure how many Mpix your photos have, but 90k key point limit coud be over the reasonable value and maybe in some case misleading for alignment algorithm... it depends on photo's resolution. In guided image matching option you can set key point limit per Mpix, so your photos would have more consistent amount of keypoints.

Did you saw some errors or strange unusual lines in console log during depth maps calculation?

45

General / Re: Agisoft Metashape Pro on Server wit NVIDIA L40s

« on: June 26, 2024, 11:15:05 AM »

It is the latest NVIDIA architecture same as desktop versions, so I see no problem.
More local network instances of Metashape make sense even with just one L40, especially detecting points phase will bennefit from large amount of instances utilizing just one L40. The only limit is the amount of VRAM, e.g. with 8GB VRAM and processing 18Mpix photos you can run ~6-7 local instances during detec. points. With L40 48GB VRAM, you can run ~36-40 local instances to amazingly speed up processing.
In other processing tasks(depth maps) it does not make much sense run so many instances and 1-3 instances per GPU would be enough.

For detecting points(36-40 local instances) you will need also high core  count CPU( 32c, better 48-64) and that is just for one L40. Also very fast SSD... 36 instances will make presure on disk reads, because several tens of photos(let's say 10MB per photo) will be read from disk per second ~500-1000MB/s  ...it depends also from CPU(speed and core counts), which is decoding all photos during this phase.

So the question is, if it make sense to utilize 4 L40's and if you will have enough CPU power to feed them with data. I see no problem in depth maps generation task, but in other tasks I would replace at least two of them with cheaper variants or with another server CPUs. In other words make the server more balanced.