Mechanism to force image to be treated as a tensor

[jleibs]

Not all 2D tensors are images. Data like an stft looks like an image but doesn't play nicely with the image viewer.

The tensor viewer has a number of nice features like:

• support for f64 data
• the ability to not lock the aspect ratio
• support for colormaps

We can trick the viewer into creating a tensor by adding a trivial outer dimension, but it would be nice not to need to do this:

Example:

f, t, Zxx = scipy.signal.stft(x, fs, nperseg=1000)
Zxx_image = np.abs(Zxx).astype(np.float32)
Zxx_tensor = np.expand_dims(np.abs(Zxx), axis=0)

rr.init("stft_test", spawn=True)
rr.log_tensor("stft_image", Zxx_image)
rr.log_tensor("stft_tensor", Zxx_tensor)

[Wumpf]

Related to

• Allow configuring the kind/category of a space view after its creation #1164

[jleibs]

rr.log("stft_tensor", rr.Tensor(Zxx_image))

now does exactly what this describes