Hello,
In order to make the Evidence-Based scale invariant, would it be enough to increase the hypothesis space with a new transformation? I mean, instead of having just rotations, could we use also scaling? Something like:
a = [x; y; z]
S = [[sx, 0, 0]; [0, sy, 0]; [0, 0, sz]]
a_scaled = S @ a = [sx * x; sy * y; sz * z]
Thanks!
Sergio
Hi @sergioval, you’ve probably already seen this, but just putting a pointer to the roadmap item for Supporting Scale Invariance for anyone else interested.
More later.
Oh! I had not, thanks for the pointer. I see this potential solution was already considered: “Testing of different scales in parallel, similar to how we test different poses of an object.”
It makes sense, cool, thanks!
If you have any more questions / ideas let us know!