Why is the ground truth depth map so sparse?

[TruongKhang]

I loaded the depth map and got the result like this:

Why is it so sparse and how can I get the dense version?

[puzzlepaint]

It looks like this image has been downsampled with nearest-neighbor filtering. This means that a lot of depth pixels are missing from this version. If you downsample with another technique that preserves that information, then it should be pretty dense at this resolution.

[jianweike]

It may be a dumb question, but how do you actually load the ground truth depth map? The depth maps I downloaded are a set of .JPG files that can not be loaded(not a JPEG file: starts with 0x18 0x0a).

[puzzlepaint]

The images are not actually JPEGs, their filenames just match the original JPEGs to indicate that they correspond to each other (in hindsight that was a bad decision since several people got mislead by that). The ETH3D documentation describes the actual format in the last paragraph of the training data description here:
https://www.eth3d.net/documentation#format-of-multi-view-data-training-data