Igor Susmelj
Igor Susmelj
I stumbled across this repo to convert a PyTorch model to tflite without introducing transpose layers everywhere. After several attempts messing around with TensorFlow, Keras, and PyTorch version combinations I...
The style guide started as a small side project for personal use. We then started using it within our company. Now, I heard from friends that some of the largest...
## Changes - add benchmark for ImageNet
We did not include the BitMask lib in the docs. Furthermore, it would be great to add some examples for common use cases. E.g. create bitmask from a tag based...
We have a model zoo but no documentation on how to load the models from it. We also have the PyTorch Lightning wrapper for training the models but no information...
Let's write a simple example on how to train a lightly embedding model using [Neuro](https://www.getneuro.ai/) Requirements of the example code: - Small dataset (max a few hundred images) - Everything...
It would be great if one could use VISSL pretrained models and work with them in Lightly. Since both frameworks use PyTorch that should not be too complicated.
Be more clear about the models and mention the difference in implementation compared to the papers
For SimCLR and MoCo there are two versions. Our implementations are more close to v2 of both from what I've seen. But I'm not sure we use all the changes...
[Mean Shift for Self-Supervised learning](https://www.csee.umbc.edu/~hpirsiav/papers/MeanShift.pdf) introduces a simple extension to BYOL. Code is available here: https://github.com/UMBCvision/MSF From paper: ![image](https://user-images.githubusercontent.com/1522766/114239554-84589f80-9986-11eb-8b5d-43819349189a.png)
From the documentation, it is not yet fully clear how to use lightly with a custom dataset and dataloader. I would add the following: - add to the docstring of...