Releases: FrancescoSaverioZuppichini/mirror
Releases · FrancescoSaverioZuppichini/mirror
Beta
Beta
New changes
- create a base class for visualisation in
mirror.visualisations.Visualisation
- new way to defined custom visualisation for the web app:
WebInterface
must be implemented to allowsmirror
to map the visualisation to the UI
- fixed GradCam on Grey images
Beta
Mirror now supports multiple inputs in booth front-end and server side.
Changes
- added a bottom tab to select the current input
- improve the cache on the server side to also take into account different inputs
Breaking changes
- you need to pass an array of inputs to
mirror
Beta
- web app
- added input text field to select class id
- visualisations
- moved all visualisations in their module so they can be used without mirror
- fixed colors in grad cam
- fixed algorithm in grad cam
- added
relu
in the output
Beta
This is the first beta of mirror, a CNN visualization tool build for PyTorch.
- Weights, Back Propr, Deep Dream and Grad Cam visualisations are available