Question: Test-time resolution scaling up

[schirrmacher]

In the paper you mentioned "test-time resolution scaling up images". I am not sure what this exactly means.

However, we surprisingly find that our model has the property of “test-time resolution scaling up”. [...] It means we can almost freely increase the image resolution at test time to produce more fine-grained depth maps.

Can I just increase the resolution when performing inference by setting the input size here or is another training required?

[schirrmacher]

Okay I found it out by myself here for the interested:

Here an inference with --input-size 518:

Here an inference with --input-size 1036:

So yes, the increase in inference input improves the quality! Great job!