Rerun ERROR: Shape must be rank 2. when logging depth image with shape (H, W, 1) from C++

[roym899]

Describe the annoyance
Minor annoyance: when logging a DepthImage with a tensor shape of length 3 (such as {height, width, 1}) Rerun reports

Rerun ERROR: Shape must be rank 2.

on every log call; however, the depth image shows up nonetheless.

This is a little annoying, because the same code to log an Image will not work when logging a depth image, i.e., you need some separate logic to get the tensor shape for a depth image compared to a normal image.

In Python there is no equivalent warning when logging a DepthImage with shape (H, W, 1).

Workaround

rerun::DepthImage({img.rows, img.cols}, rerun::TensorBuffer::u16(img))

instead of

rerun::DepthImage({img.rows, img.cols, img.channels()}, rerun::TensorBuffer::u16(img))

Expected behavior
Accepts tensor shapes of rank 3 as long as the last dimension has size 1.

[roym899]

Related to that, I think it'd be easier to understand the error if it says "Shape must be length 2." instead of "rank 2". I think rank isn't even a well-defined term for a vector.

[emilk]

I agree that "rank" is a confusing term. I suggest we replace "rank N" with "N-dimensional" in our language. In this case, the error message should be something like "The tensor must be 2-dimensional, but got shape 800x600x1".

Of course, the problem with that is we need to do string formatting in C++, which is hell.