MalteEbner

I solved a similar error by switching to python < 3.9: https://github.com/heartexlabs/label-studio/issues/1285

The solution that worked for me on Linux: The docker requires access to the `cuda` library during build time. To ensure this, make sure that your `/etc/docker/daemon.json` file looks as...

Would it be possible to already have a basic typehinting functionality for types like 'float', 'dict'... that just gives 'object' as a type hint instead of importing the more precise...

> > Would it be possible to already have a basic typehinting functionality for types like 'float', 'dict'... that just gives 'object' as a type hint instead of importing the...

@Dringho Do you know why this PR is not approved? Does it need to be rebased on master or are there additional problems?

Hey @dreamercv , ### Regarding hyperparameters: These benchmarks show that usually, more epochs and higher batch sizes are better: https://docs.lightly.ai/getting_started/benchmarks.html We did not benchmark different optimisers so far, best take...

Most SSL models learn to distinguish images, thus their embeddings encode properties of the full image and neglect details. This makes SSL pretraining great for classification tasks, but worse for...

We already make sure that we don't upload embedding files not matching the specs, as they cause the reordering according to the mapping to fail (see below). However, you are...

It might not be a big problem, but I found that we don't have a CLI keyword for uploading metadata . Instead, if `upload` is neither `thumbnails` nor `full` ,...

I agree that using the standard models offered in torchvision would be better than having our own models. What we are currently thinking about is a good migration strategy, as...