Laszlo Rikker

Hi @edgarriba, just found this thread, I've started looking into this and make it jit compatible. Got stuck on the "CoarseMatching" has no attribute 'log_optimal_transport' - line 135 - coarse_matching.py...

Hello @edgarriba , I think I got to the end of this now (currently on loftr.py) and managed to map most of it to be jit aligned. I'm still getting...

hi @edgarriba, just did that. could you have a look when you have some time? i removed/refactored some parts of the code that adds complications and is not used in...

Running 'TestLoFTR.test_jit()' will result in the following: `RuntimeError: Unsupported value kind: Tensor`

i also attempted to serialize each individual modules separately to see which caused an error. the only one I managed successfully serialize and then load was the "backbone".

@edgarriba @ducha-aiki, when i run jit.trace i get the following: `RuntimeError: 0INTERNAL ASSERT FAILED at "C:\\actions-runner\\_work\\pytorch\\pytorch\\builder\\windows\\pytorch\\torch\\csrc\\jit\\ir\\alias_analysis.cpp":607, please report a bug to PyTorch. We don't have an op for aten::fill_ but...

So far the following modules are jit scriptable; _backbone, LocalFeatureTransformer_ and _FinePreprocess_. The _PositionEncodingSine_ module has got a "register_buffer" call which complicates things. The _Coarse/FineMatching_ modules both have dict "updates"...