Philipp Wirth

That's strange, I'm not able to reproduce this with: ``` lightly==1.2.13 prefetch-generator==1.0.1 ``` `lightly-embed`: ``` > lightly-embed input_dir=/datasets/cifar10/test/frog/ Compute efficiency: 0.23: 100%|███████████████████████████| 1000/1000 [00:01

Is it possible that you need to restart the runtime? I tried to reproduce the issue here: https://colab.research.google.com/drive/1AhoCdvetvPo0IhWTw90GV2O2MUq2EyDm?usp=sharing Or maybe the error happens at a different place now? In case...

Hi @albertoRodriguezVW You are exactly right. The selection is still the one done on embeddings A. It's simply visualized using embeddings B. Thanks for the feedback! We'll try to make...

> Sorry for the typos. > > 1. The prediction head is added on f_\theta. > > 2. argmin -> argmax in Eq.4. > > 3. SimCLR's augmentation is OK...

Current best result: ``` ------------------------------------------------------------------------------------------ │ | Model | Batch Size | Epochs | KNN Test Accuracy | Time | Peak GPU Usage | │ ------------------------------------------------------------------------------------------ │ | SMoG |...

Can be solved by this: https://stackoverflow.com/questions/38361505/how-do-i-locate-python-command-line-tools-after-installing-to-non-root-user-ac Let's add a note to the docs 🙂

A good image for the active learning page would be this: 

I agree that we should offer such a high-level interface. I made an example based on [this](https://huggingface.co/transformers/model_doc/roberta.html#transformers.RobertaConfig): ```python from lightly import SimCLRConfig, SimCLREncoder # initialize configuration (has all information about...

The idea behind this was to have several layers of abstraction. A user who wants to access the library on a lower level might want to use a model with...

Yes, it is. However, parts may change with the ongoing refactoring so I'd recommend to work on something else atm.