Add Weights and Baises Integration

[AyushExel]

This PR adds support for Weights and Biases Metric and evaluation logging.

How to use?

• pass --use-wandb flag with tools/train.py
• Set number of interactive evaluations samples to be logged at each validation step by setting --eval-samples <int>

Tables

Tables Inspect, filter, query, and compare your model prediction to better debug

In the current implementation, the wandb arguments are directly passed in the Base Logger. We can refactor it out into a more suitable place if needed. There are more points of integration such as model checkpointing on cloud, interactive inference logging which can be built on top of this

[RangiLyu]

Thanks for your contribution! I'll review this PR as soon as possible!

[RangiLyu]

In the current implementation, the wandb arguments are directly passed in the Base Logger. We can refactor it out into a more suitable place if needed.

I think you can create a new class called WandbLogger and provide the same api as the NanoDetLightningLogger.

And currently, the logging logic in task.py is not designed for other loggers (e.g. I used tensorboard's add_scalars api in task.py ).

You can provide a draft code and I'll refactor base on it.

[AyushExel]

@RangiLyu I'm using the baselogger's API to create a new WandbLogger. I'll make the task accept a list of loggers so users can implement and add their own loggers later on.
Hopefully the only change needed weill be to replace
self.logger.xxx
to

for logger in self.loggers:
     logger.xxx

[AyushExel]

@RangiLyu hey, any updates on this?

[RangiLyu]

@RangiLyu hey, any updates on this?

I'll create another PR based on this. Some refactoring and unit tests should be added. I'll do it this week.

[AyushExel]

Hey, is there a blocker here or something that we can help with?