Open tonetechnician opened 9 months ago
No i reimplemented it with some minor changes as well since their code is really hard to retrain
Hello! I also have some questions about the LANet port and original usage, I hope you don't mind me jumping in this issue thread.
In your comparisons did you test the original LANet? I am struggling to run it at any sensible speed like their few seconds inference time advertised. My tensorflow-gpu version is correctly installed but my GPU usage only ever goes up to 3% with a 3090.
Did you face any similar issues when testing? Any tips would be useful.
as far as i remember it was running for testing but it was totally unusable for training
Can you remember what sort of inference speeds you were getting?
it was noy realtime, i think it took about an hour. then i figured i have to reimplement
Hi there,
Thanks for sharing this interesting work! I'm particularly interested in the pytorch port of LANet as the original source code is quite dated. I wondered if you retrained your port or if you were able to port the tensorflow weights from the original repo?