Agisoft Metashape 1.8.0 pre-release

[Paulo]

I am trying to check if there was really a zoom factor looking at the distances calculated after alignment and scaling with cm scale I get average distance of about 19 cm ....

[bgreenstone]

I think those test photos were from an iPhone X.  Not sure if he used the 2x or normal lens.  If it’s the normal wide angle lens then 19cm is about right.

[Paulo]

Hello ,

I have found that in 1.8 pre release, when I place shapes with labels in the chunk and  save project. Then when I reopen the project, the shapes appear but without their labels.
In first scrren, I show 2 linestrings placed with their label equal to the line length.
When I save and reopen project, same 2 linestrings appear without their labels (2nd screen copy).

The shape label is however well present inside the doc.xml from the shapes.zip compressed folder from the <project_name>.files folder structure as seen in:

</shape>
    <shape id="1" group_id="0" type="polyline" label="0.486 m" has_z="true">
      <vertices>-1.1511491242687065e+02 3.5940266001437223e+01 8.1961011157548512e+02 -1.1511491430060157e+02 3.5940268773995314e+01 8.1927352960266114e+02</vertices>
      <meta>
        <property name="Dist_CC_PP" value="0.48635831257152956"/>
        <property name="Photo" value="20211220_214523866_iOS.jpeg"/>
      </meta>
    </shape>

[Alexey Pasumansky]

Hello Paul,

Which pre-release build you are using and if you are assigning labels via Python API or manual input via GUI?

[Paulo]

Hi Alexey,

I am using latest release 1.8.0 build 13736 and shapes were placed with Python API.... if I open same project in 1.7, labels are visible....see following

[bgreenstone]

The new filtering functions have certainly helped with 98% of the giant blobs of extraneous geometry that earlier versions put underneath meshes, but the new version still generates a lot of this stuff in every scan.  It's almost always tiny, disconnected blobs so they're easily removed in Blender, but still, they shouldn't be there, and in 1.7.3 and earlier you never get these. 

In this screenshot from blender you can see all of the small blobs of geometry in orange.

[bgreenstone]

I just ran 2 batches and both had pretty bad blobs on the underside.  One in particular had them *really* bad - it's unusable geometry, so I'll have to process both of these in 1.7.3 for now.

[wojtek]

chunk.model.cropSelection() executed from python deletes entire Undo history in the latest 1.8.0 build with the new Undo list.

[Alexey Pasumansky]

I just ran 2 batches and both had pretty bad blobs on the underside.  One in particular had them *really* bad - it's unusable geometry, so I'll have to process both of these in 1.7.3 for now.

Hello Brian,

We need a project data to reproduce the problem, ideally camera alignment results and source images, so that we could run the depth maps based mesh reconstruction procedure on our side.

As for the "tiny blobs", you can remove them using Main Menu -> Gradual Selection -> Connected Components criterion, but a sample project would be also helpful for additional investigation.

[bgreenstone]

Unfortunately, there’s no way to do Gradual Selection in a batch

[bgreenstone]

Sorry for always being the bearer of bad news, but it seems every day I encounter a new issue with 1.8 that works perfectly in 1.7.3.

Here is a new issue that came up yesterday on several objects that I tried processing from one customer.  First some explanation what we are doing:  We make 3D printed display bases.  The customer makes an impression mold of the object they want a base made for.  Then they photograph the mold and upload the photos to us where we process them in Metashape and build their display base in Blender.  That's why in all of my samples you see a box of sand and a ruler.  The ruler is for scale, and the box of sand is the impression mold the customer makes.

In this case, the customer used a featureless white box to hold the sand (I don't know why his photos are all tinted pink).  Metashape obviously cannot properly scan the blank white surface, so that geometry is never reproduced in any version of Metashape.  No big deal since we don't care about the box anyway.  However, in 1.9 the area of the box that didn't scan is causing the ruler nearby to have the texture map incorrectly generated.  In these photos you can see that it came out perfectly in 1.7.3, but with 1.8 there is some incorrect texture overlapping the ruler.

I have uploaded the entire project with a Chunk done in 1.9 and a Chunk done with 1.7.3 so you can easily compare.  I have also included the Batch file I used in each version.  The Batches are identical.

http://gofile.me/61Yhd/Kea9kcby7

[bgreenstone]

Sorry, but another bug to report.  I normally do Orthophoto texture map generation at 8192x8192 for my scans.  Today I decided to try a Generic texture instead, and it froze up 73% of the way in.  The only way out is to Force-Quit Metashape.  I tried a 2nd time and the same thing happened but at 74%. 

Of course when I do it with 1.7.3 it works perfectly.   Only freezes up on 1.8

I have uploaded the test project here:  http://gofile.me/61Yhd/xzM4z3TlE

[Alexey Pasumansky]

In this case, the customer used a featureless white box to hold the sand (I don't know why his photos are all tinted pink).  Metashape obviously cannot properly scan the blank white surface, so that geometry is never reproduced in any version of Metashape.  No big deal since we don't care about the box anyway.  However, in 1.9 the area of the box that didn't scan is causing the ruler nearby to have the texture map incorrectly generated.  In these photos you can see that it came out perfectly in 1.7.3, but with 1.8 there is some incorrect texture overlapping the ruler.

Hello Brian,

Projection plane for the orthophoto texture mapping is selected differently for these models, since the distribution of the mesh faces is different. For the 1.8.0 project the orthophoto plane is slightly tilted relatively to the table's surface, therefore the average blending mode mixes the colors from the images of the ruler and the box in certain area.


As for the second issue - freezing of the Generic mapping mode, we are currently working on the fix.

[bgreenstone]

Thanks, Alexey!  I've always wondered how the orothophoto plane is determined.  How does Metashape determine where the Z-axis is to project the texture along?  I've done about 500 of these things, and previously it always managed to correctly project straight down the object's z-axis.  This is the first time it hasn't, and I don't know what is different that would cause it.  Also, why did it project correctly in 1.7.3? 

-Brian

[Alexey Pasumansky]

Hello Brian,

I have cropped a part of the model from 1.8 chunk and then rebuild the texture using Orthophoto mapping option and it produced the expected result.

The mesh in 1.7 chunk is bigger (due to bigger bounding box), so the distribution of mesh faces is different and it allows for proper selection of the orthophoto mapping projection plane. If you try to build the texture in 1.8 version using the model from 1.7, you should get similar result, as you have in 1.7.