Recent Posts

1

General / Re: Lines in Dense Cloud

« Last post by dcobra on Today at 01:49:31 AM »

Mak,

Thanks for the suggestion.  There doesn't appear to be any white lines in the images or any errors with the regenerated texture.  See dense cloud and texture attached.

2

Other Languages / Re: Вопрос по Vector в python

« Last post by godszerg on January 18, 2019, 10:30:19 PM »

Далее вопрос в том, какая задача стоит перед Вами. Если хочется, зная локальные координаты, создать маркер, то это можно сделать так:

Code: [Select]

T = chunk.transform.matrix
chunk.addMarker(T.inv().mulp(Metashape.Vector([1.6, -7.7, 2.2])))

А можно как то всю систему координат чанка инвертировать?

3

Other Languages / Re: Вопрос по Vector в python

« Last post by godszerg on January 18, 2019, 10:18:45 PM »

Спасибо за ответ! Задача такая. Хочу наладить связь между проектом в Unity 3D и Metashape. Координаты из Unity передать в Metashpe и зменить допустим размер и положение (наклон в том числе) region (bounding box)

Для проверки я добавил маркер в ручную в программе с координатами из юнити и положение маркера совпало с точкой в юнити. А вот если я добавил маркер через скрипт в этот же проект получил результат выше и другое положение маркера.

4

General / Re: GCPs Have High Error Values

« Last post by egross on January 18, 2019, 09:30:09 PM »

Hello JasonR and Alexey,

Thanks for your replies!

Are your targets unique in the sense that you didn't accidentally pick the wrong one when marking them? Are you optimizing your cameras with the GCPs selected? Those are some common errors to check.

I double checked my targets and I didn't accidentally make any wrong picks necessarily. However, I must admit that while capturing the images, I did not have GCP targets placed all around the field site. Bushes and small, unique-looking rocks were essentially used as my GCP "targets". I feel that I picked these bush/rock targets quite accurately. Before posting here, I did unchecked my cameras and only selected the GCPs when I did the "Optimize cameras" option as that combo seemed to improve my accuracy.

Hello egross,

Are you observing any high projection errors (in pixels) for any of markers? Also is this table shown before or after optimization operation, and if it is after the optimization, which coefficients have been adjusted?

Please see the attached screenshot for the projection errors associated with each marker. This is my first time attempting to employ GCPs so I'm not too sure if the projection errors (in pixels) are too large. Both tables that I have attached in my forum posts associated with this topic show results after the optimization operation. For the optimization operation, I only selected "Adaptive camera model fitting" (Selecting this option grayed out all of the options listed in the General section of the Optimize Camera Alignment dialog box).

5

General / Re: Apply masks to tie points not working

« Last post by eca4cfc on January 18, 2019, 09:18:35 PM »

Hello,

I just attempted the photo set again with both the generic preselection method selected and unselected. With generic preselection disabled 41 of the 46 photos could be aligned; however, the photos of the bottom of the object either did not align properly or could not be aligned at all. The background was also still visible within the sparse cloud. The number of cameras and points is slightly higher with the method disabled, but it is still not correctly aligning.

I also tried with the banana dataset with generic preselection disabled and the photos all aligned properly.

Any ideas on what could be going wrong with the align photos process? I've linked the photo set below.

https://drive.google.com/open?id=14XH_rDuHLt250yu5dZpRYfTqQQfj-apw

6

Other Languages / Re: Вопрос по Vector в python

« Last post by Alexey Pasumansky on January 18, 2019, 09:00:41 PM »

Добрый день,

Дело в том, что практически все координатные элементы, соответствующие содержимому чанка, имеют координаты во внутренней системе, которая отличается от координат итоговой модели, если чанк привязан или даже отмасштабирован.

Речь идёт, например, о таких элементах, как camera.center, marker.position, region.center и т.д.

Функция addMarker также на вход получает координаты в этой внутренней системе. Если чанк в вашем проекте привязан в локальных координатах, то связь между внутренней и "внешней" системой такая:

T = chunk.transform.matrix
coord_external = T.mulp(coord_internal)
в обратную сторону:


Далее вопрос в том, какая задача стоит перед Вами. Если хочется, зная локальные координаты, создать маркер, то это можно сделать так:

Code: [Select]

T = chunk.transform.matrix
chunk.addMarker(T.inv().mulp(Metashape.Vector([1.6, -7.7, 2.2])))

7

General / Reprojection error treshold

« Last post by Tchale on January 18, 2019, 09:00:23 PM »

Hi All,
I am working on a project and I am getting an RE of 0.395. I have tried the "gradual selection" tool but then I got holes in my map. I am wondering if 0.395 is acceptable; this because most tutorials I have seen state that the RE threshold  should be 0.3.

Thanks.

8

General / What's the best workflow for 300mil mesh having only 32gb of RAM?

« Last post by kirk on January 18, 2019, 06:13:24 PM »

I need to build a super hi res mesh for a terrain surface  from Mavic Pro drone dng images   to be then baked as displacement/height /normal map textures of a shell low-poly one.   

Looks like main limitation is RAM, right?   Is there a way to work around it?     I once build 800 mil mesh on the same pc  but it's kind of foggy what settings I had that time.

 I need maximum possible details in the target normal texture so I bet only way is to build a super huge mesh.


1.  I should go "Heighfield" right?   Not "Arbitrary"    So the new depth map mode is not an option, right?   I tired "depth map" and it stops at 97% decimation stage and never finishes it.

2.  What should I set for "reuse depth map", does it matter at all?

3. I shouldn't even try "decimate" , right ?      It doesn't work with huge meshes?
 I actually somehow coped with  "simplifing" my huge meshes in a Steam version of Reality Capture.   Unfortunately  RC itself fails with proper Mavic Pro images  aligning and even when it doesn't,  it does lots of surface artifacts over what Agisoft does  and also inferior/ more blurry color textures IMO (Needs much more hi res input than the drone  is able to provide considering flight/battery time)

4. What's best geometry format to export from Photoscan . Takes less time to save/ less GB on disc  etc.?    Alembic?

5.  Would a color texture  build for extra  low poly mesh with custom unwrap  be matching  height/displacment details baked from super hi res mesh in 3d party texture baker?    I mean considering texture bakers usually do ray tracing along low poly mesh normals?  Would it match with how Agisoft projects the texture pixels?

 Or I should make the color texture for same super hi poly object too?  Would take forever I guess

ps.      Could sombody who use drones share few thoughts.  Is it good idea at al or anything I am going to get is lots of geometry noise instead of super hi res mesh?     

9

General / Re: Depth Maps for newly aligned photos?

« Last post by Alexey Pasumansky on January 18, 2019, 06:11:57 PM »

Hello MitchellT,

We have made a comparison test for your project for the mesh based on the High quality depth maps (re-using the alignment from your original project) on the following configurations:
Intel i7 3960X + GTX 680
and
AMD Ryzen Threadripper 2950X + 2x GTX 1080 Ti

The total time for the Build Mesh stage (not including the depth maps calculation) for the first configuration was about 10 hours 30 minutes and for the second - 2 hours and 5 minutes. From this time CPU only part in the beginning and in the end of the job took 1.5 hours for the first configuration and 1 hour for the second one. So the part that uses GPU quite intensively took about 9 hours on GTX 680 and 1 hour on 2x GTX 1080 Ti. In both cases "use CPU" flag has been disabled.

The issue with the memory that you were observing in the log should be fixed in the next update.

10

General / Re: DEM creation glitchy, ortho window slow

« Last post by PwnageGIS on January 18, 2019, 06:05:29 PM »

Yes! That fixed my problem! Thank you so much Alexey!

I didn't notice that there were issues in the Photo view navigation either, since I have not needed to create masks or marker for my DEM's or Orthophotos. I thought I had set my drivers to automatic install for new updates, but I guess not. Thanks again!