It doesn't work well for my custom datasets.

[bring728]

First of all, thank you for sharing your great research.

I have 9 images and the cameras are all facing the same direction. I want to get 9 depth maps.
For the DTU dataset, depth estimation works fine with your network.
But when I input my image, I get the following bad result.

I don't know if it has anything to do with it, but the camera coordinate system I used is right, down, forward. And with depth_min = 1.25 I set depth_inverval=0.1.
In the case of num_view, it is set to 9 so that all images are always used.

Do you have any guesses why?

[bring728]

Oh, I made a mistake. There was a part where the intrinsics matrix was divided by 4, but I omitted it. I solved it, and the result is good to some extent.

[Sun-Xinnnnn]

How did you solve this problem? I had a similar problem......

[TruongKhang]

@zhang-snowy, if you use our model for your custom dataset. We recommend you read the data processing files (dtu_yao.py, blended_dataset.py) as well as the structure of datasets in detail. and then prepare your data in the same way