Enhance single channel denoiser and WaveformNode - #4673
Conversation
chrishalcrow
added
the
Edinburgh hackathon 2026
| spike_size = waveform_node.nbefore + waveform_node.nafter | ||
| # Load model | ||
| self.denoiser = self.load_model() | ||
| self.denoiser, model_relative_path = self.load_model( |
There was a problem hiding this comment.
Move it to a separate function and have the node accept the instantiated model
| parents: Optional[List[PipelineNode]] = None, | ||
| model_folder: Optional[str] = None, | ||
| repo_id: Optional[str] = None, | ||
| model_name: Optional[str] = None, |
There was a problem hiding this comment.
We should be able to give the model directly in this signature and skip the autoload when we want.
|
OK for me but for noes we should always have the possibility to give the model directly and skip the laod mechanism |
then you cannod check the model_info. The idea is you can pass a fodler with the weights and the model_info.json. Isn't that enough? |
|
If the model is given we can skip tests. |
|
@samuelgarcia done in last commit. Also when you pass an external model it checks the last layer dimentions |
3 participants
Extend flexibility of
WaveformNodeand single channel denoiser. Now testing against toy and yass models here: https://huggingface.co/SpikeInterface/waveform_denoiser