catch 22 need masks to align photos but initial align fails

[Steve003]

Hi.
metashape 1.5.3

fundamental catch 22.

1) align photos, 3 chunks because object 3 positions relative to background, get an overall 43% failure rate.
2) it was then to be dense point cloud then mesh then edit background out then mask from model then merge chunks then realign photos,

will all that work so as to get mask from model with almost half failed ? model wont be  good., mask wont be good.

catch 22 you  need mask from model to then re-align well yet unable to get initial photos to align without masks !

I see that David Cockey uses this method, any help David or anyone appreciated.

update ....went ahead and edited out background, import mask from model, 21 masks appear from 54 photos, of those they are all chunk 1, chunks 2 (18 pics) and 3 (8pics) none.

Steve

[Alexey Pasumansky]

Hello Steve,

If each chunk represent the fixed object-to-background relative orientation, then the images of the same sub-set should properly align, providing that there's sufficient overlap between the images.

Suggested masking from the background would help to align images from different sub-sets together in the single chunk.

[Alexey Pasumansky]

Masking from model, if applied to Entire Workspace, would generate masks for the images from the certain chunk basing on the model from the same chunk. It is not possible to generate the masks for the images from the "Chunk A" basing on the model from the "Chunk B".

[Steve003]

Hi Alexei,

If each chunk represent the fixed object-to-background relative orientation, then the images of the same sub-set should properly align, providing that there's sufficient overlap between the images.

quite right and I have sent you the project file and images, as you will see they overlap well. 1 chunk per object fixed relative to background.

Item sitting on the gridded board on concrete driveway in a yard, I moved around it. one of many projects and photo sessions like that. No chance of a background image or two or 3 representing entire background. the object predominates the photo so of the area of it photographed it should make a model, but if they fail to align I cant even go ahead and make dense point cloud/depth map and then make the model from which to edit the background from, then import mask from.

pics that dont align at the outset, 24 out of 54 failing,  thats 'game over'.

Masking from model, if applied to Entire Workspace, would generate masks for the images from the certain chunk basing on the model from the same chunk.

Are you saying I should choose 'Entire Workspace'  as opposed to all cameras when importing masks from model for chunk 1 ?

I havent got as far as that yet because I havent even made dense point cloud or mask,  I dont have decent models to remove background from yet. I have only aligned and with so many unaligned I cannot progress.

It is not possible to generate the masks for the images from the "Chunk A" basing on the model from the "Chunk B".

Agree wouldnt have crossed my mind.

first step is to align pics having added photos to project and they wont align, I can accept 1 or 2 failures but not 24 out of 54.

I then try photoscan 1.2.6 and get almost half the failures. I have sent an email with differences.

Should not metashape better photoscan ? The settings clearly make a difference.

I used:-
Accuracy: HIGH
Pair preselection: DISABLED
Key point limit: 40,000
Tie Point Limit: 10,000
Constrain features by mask: greyed out unselectable
Adaptive Camera Model Fitting :  ON


try my images in photoscan 1.2.6 to see what I mean.

Regards

Steve