pytorch-lightning

🐛 Bug

Removed _convert_params function in new pytorch-lightning release (1.6.0).

To reproduce

from aim.pytorch_lightning import AimLogger

trainer = pl.Trainer(logger=AimLogger(experiment='experiment_name'))
trainer.logger.log_hyperparams(hparams)

Expected behavior

• Successfully log the hyperparameters with no errors

Environment

• Aim Version (3.8

🐛 Bug

found another paper-cut with instance segmentation...
it is not trivial/intuitive how to match mask with input image as output can be any size and mask is 128x128
so it is just scaled equally in each dimension is applied padding uniformly?

To Reproduce

https://www.kaggle.com/jirkaborovec/cell-instance-segm-lightning-flash#Training-with-Flash-Lightning

Additional cont

"Multi-task training to efficiently combine object detection, segmentation, and classification models" is a feature of icevision. However, I don't know how to use icevision to do multi-task learning like object detection and classication. I just need a simple example without details.

Don't remove
Main issue for examples: #39

Describe the bug
Not necessarily a bug. The dependency of pandas == 1.1.5 is a really obsolete version, (and also many other packages)
For pandas I read through the source and didn't find a version specific usage. Considering change it to pandas >=1.1.5?

Add a function and module that permits spike-time decoding, as suggested by @schmitts https://twitter.com/sbstnschmtthd/status/1432343373072019461

It is always good to have more options to choose. So it would be a good idea to add more schedulers. The steps are the following:

• in conf/scheduler add a config for a new scheduler
• if this scheduler requires some other library, update requirements
• run tests to check that everything works

Example: https://github.com/Erlemar/pytorch_tempest/blob/master/conf/scheduler/cyclic.yaml
`