Large dataset stitching issue

[Zeeshan97]

Hi,
I have encountered a particular issue while working with a dataset consisting of 3100 RGB images covering an area of approximately 232 acres. Specifically, I have noticed a discrepancy in the resulting orthophotos when attempting to stitch different subsets of images.

When stitching together smaller subsets, such as 500, 100, 2000, and 2500 images, the resulting PNG orthophotos appear to be of satisfactory quality. However, when attempting to stitch all the images together to create a complete orthophoto, an abrupt transition or irregularity becomes apparent.

Here are the screenshots of the results obtained when stitching together different subsets of images from my dataset:

The subset of 3000 images:

I have considered the possibility that the observed issue in the orthophotos could be attributed to sun reflection in water or other shadow-related factors. Could this be the reason?

I am using a Core i7-12th Gen processor and 128GB of RAM. The parameters I am using with nodeODM are:

dsm: true
dtm: true
orthophoto-resolution: 5
fast-orthophoto: true
orthophoto-png: true
ignore-gsd: true
sfm-algorithm: plannar
tiles: true

[Zeeshan97]

The subset of 2000 images:

[Zeeshan97]

The subset of 2500 images:

[ThorZ]

It would be helpful if you can attach the processing logs.

[Zeeshan97]

I am using the nodeODM. But I have written a Python script through which i can write down the processing log. I’ll find and attach the file

[Zeeshan97]

This image is the result of dataset images number 2501-3000.

Can there be any problem in the raw images of this stitched which cause issue when the ODM ran on complete 3000 images?

[Saijin_Naib]

Try processing all of your images at Default before you go tuning settings.

For instance, do not ever use --ignore-gsd.