Apologize, it is caused by the poor amount of tie points (4549 by high - fail; 46646 by highest - success), not caused by the duplicate image names.
Please archive or delete this post if necessary.
I am using "Foldio 360" rotation platform to take object images for 3D reconstruction, it will rotate 10 degrees (horizontally) each time and take 36 images, and always name as follows: [01.jpg, 02.jpg, 03.jpg, ... , 36.jpg]
The camera position is always fixed, but scanned object is flipped for 45 degrees (vertically), and finally got 8 image groups:
/root/
|- rotation 0/
| |- 01.jpg
| |- 02.jpg
| |- ...
| |- 36.jpg
|- rotation 45/
| |- 01.jpg
| |- ...
| |- 36.jpg
|- rotation 90/
|- rotation 135/
|- ...
|- rotation 315/
(please note, for each folder, the subimages have the same image name)
Then, I added all these images in one chunk, and grouped them by camera groups, and make masks for each image to remove the static backgrounds.
Successfully aligned photos by [highest accuracy, generic preselection, sequential reference preselection, apply mask to key points, exclude stationary tie points and adaptive camera model fitting], got expected tie points (left subimg). But build dense cloud, it produces point cloud with a missing part (right subimg), which is almost the same result from only 36 images from "rotation 90" folder.
Then I copied and rename the project folder like this, given unique image names:
/root/
|- rotation 0/
| |- c0_01.jpg
| |- c0_02.jpg
| |- ...
| |- c0_36.jpg
|- rotation 45/
| |- c45_01.jpg
| |- ...
| |- c45_36.jpg
|- rotation 90/
|- rotation 135/
|- ...
|- rotation 315/
Doing the same thing like before (except ranaming image.label c0_01 -> 01 for sequential reference preselection by script in Metashape project, otherwise unable to align all images at all), and got expected dense cloud:
The only difference between two projects, is whether image names are unique or not. I personally guess it is caused by a bug in the MVS steps of building dense cloud, the duplicate image name override previous calculation results. A simple compairson: