Álvaro Barbero Jiménez
Álvaro Barbero Jiménez
Hi there, Has anyone tried running neural-style in one of the Amazon's fancy p3 instances with Volta 100 GPUs? I usually run this on a p2 instance (K80 GPU) without...
Numpy is used only for the functions linspace and prod, but installing it requires quite a few packages that make the docker image heavier. The project could benefit from replacing...
It seems that for some images transparent patches appear out of nowhere in chen-schmidt-inverse algorithm.
Try adding the "Universal Style Transfer" method, which promises better results than chen-schmidt at fast speed. * https://github.com/sunshineatnoon/PytorchWCT * https://arxiv.org/pdf/1705.08086.pdf
It might be more easy to maintain to use the pytorch version of Gatys method, available as a notebook here: https://github.com/leongatys/PytorchNeuralStyleTransfer This also might (or might not) help running the...
Quality and stability of results can be improved by applying a color histogram transfer from the style image to the content image, prior to running the style-transfer method. Alternatively the...
proxTV does not support Windows. We should extend the CI process to create python wheels in Windows. This should be doable through Appveyor.
The python demos run but the resultant images are not being rendered correctly.
## Bug description After training a DeBERTa-v2 model with LoRA, saving the model to disk and loading it back, the loaded model doesn't produce the same output as the original...
**Describe the issue**: Reading the documentation of [`make_meta`](https://docs.dask.org/en/stable/generated/dask.dataframe.utils.make_meta.html) it states that > This method creates meta-data based on the type of x so my understanding is that a new object...