Very poor depth map, what can I do?

[JzHuai0108]

Dear authors,

Thanks for open sourcing your code.
I tried to test the pretrained weight on the train dataset 20 GB, because the test dataset 120GB too expensive for me.
I edited the dtu.yaml IMG_WIDTH 640, IMG_HEIGHT 512, and put the data in proper folders.
Then I run the command

 python fastmvsnet/test.py --cfg configs/dtu.yaml TEST.WEIGHT outputs/pretrained.pth

The resulting depth points at flow2pts.xyz are so poor that I can't see anything meaningful.
For instance, scan1/0000000_flow2pts.xyz

The corresponding image is

Is there anything I can do to improve the results?