Lukas Voegtle

icon company nvidia icon location Freiburg icon location Software Design/Architecture. Deep Learning. Machine Learning. Computer Science MSc at University of Freiburg.

Results 6 comments of Lukas Voegtle

Instructions for installing PyTorch

Probably a follow-up issue on the second option: If I do `poetry install` on that, I get ``` Installing dependencies from lock file Package operations: 50 installs, 0 updates, 0...

8.4.0 release is broken due to missing files

A quick fix for everybody facing the same issue, manually downgrade to `8.3.0`: ``` pip install --force-reinstall tenacity==8.3.0 ```

mAP is wrong if all scores are equal (=not providing a score)

Hi @MiXaiLL76 , I am sorry for not responding, was out a bit longer. It is still relevant for us, thank you very much for the quick response and even...

mAP is wrong if all scores are equal (=not providing a score)

Ah sorry, I didn't make my point clear enough: One of those variants should be correct. But we cannot settle on the variant which is the correct one. Also referencing...

mAP is wrong if all scores are equal (=not providing a score)

Thanks for this opinion! I agree in general, that the mAP of a model which over-predicts boxes but with scores is not easily comparable to a model which does not...

mAP is wrong if all scores are equal (=not providing a score)

Sorry for not responding for so long. Too much other tasks. Anyways, we added discussion in the paper: https://arxiv.org/pdf/2502.04223 Appendix S5 Effectively, my personal conclusion is, that the mAP is...