Just to inform you, those classes won’t be written into the model. You will be only able to texture the mesh using the classes.
Can you be more specific and show me some print screens? Is each layer a specific color?
What do you mean by this:
i have 104 labels
with multi classes around 30 in total
Are those classes shown on more images?
What is your exact workflow there?
How are your labels defined? Does each image have only one its label layer (with all the classes)?
Hi, can you answer also my other question?
How many classes do you have on your label images? Do those have the same color for the same class? Is the border of the class clear on the image?
In total there are 30 classes
- and on each image there are upto 10
- each class has been assigned the same color
- the borders of classes inside image is clear
this is the detailed workflow
I ran SAM to auto-segment every image, then used OPENAI CLIP to label each mask with ~30 construction classes (crane, gravel, scaffolding, …).
All masks were filtered, duplicates merged (largest‐area kept), renamed according to the RC requirements, and forced to single-channel 24-bit.
and the issue i am facing right now is that after following these steps 1) aligning images, 2) creating mesh with normal detail 3) Texturising the model.
when i try to import all the labels nothing shows up
on the other side when i only upload the labels individually it does show the classification but as soon as i increase the number of input label images
the classification starts to disappear
What are you doing after the labels import? Have you validated them?
Is it possible to share your data with us?
I HAVE ALSO VALIDATED THE LABELS.
I CAN SHARE DATA WITH YOU .
OK then, I’ve sent you the invitation for the data upload. Please add there your images, labels and project where this is happening.
Hello Abmazz,
thank you for your data. I’ve checked them and the issue is, that you are using the black color as a base and this is considered as a layer. As you don’t have labelled all parts over the all labels, there is a overlap between the black and colored part:
And as the black color is over most of the images, you are getting the grey model.
This is how it looks using one label image:
what is the possible solution of it ?
i have tried to remove the black background by making it transparent
but the issue persists
The only option here is to have all classes which are visible on the image set. As you can see on the previously attached image, you are missing yellow class on the second images. This is basically not allowed and black part takes control over that area, as there is an overlap and you have more black parts over your data.
@OndrejTrhan i have successfully classified the model, thank you for the help
there is one more thing i want to ask from you is that
which is the best way to export this classified model
i have tried exporting it directly to the cesium but it did not export the classified model
Hello @Abmaaz
The only option there is to use the classes to colorize the mesh and export such colorized model.