Adding support for wandb

[manangoel99]

This PR fixes issues #1030 and #1049 . It adds support for the wandb logger.
I've added a new command line argument --logger which accepts "tensorboard" and "wandb" as arguments and raises an error otherwise.
More wandb related arguments like project, entity etc. can be provided in the CLI using the prefix wandb-. For example

python tools/train.py -n yolox-s -d 8 -b 64 --fp16 -o [--cache] --logger wandb wandb-project <project name>
                         yolox-m
                         yolox-l
                         yolox-x

All the metrics are logged to the W&B dashboard along with the hyperparameters. The model checkpoints are stored as wandb artifacts along with tagging the best and latest models appropriately.

This dashboard uses yolox-tiny.

[CLAassistant]

All committers have signed the CLA.

[FateScript]

Thanks for your contribution @manangoel99 Plz lint your code to pass the workflow.

[manangoel99]

Thanks for your contribution @manangoel99 Plz lint your code to pass the workflow.

Fixed thanks!

[FateScript]

PTAL @manangoel99 , please also use isort=4.3.21 to sort import of your code, thanks a lot.

[FateScript]

LGTM

[samithaFHSS]

I'm getting an error when using wandb with custom data. Error is 'VOCDetection' object has no attribute 'cats'. Any solution for this?

[scottire]

hi @samithaFHSS, have you been able to resolve the above issue? If not, can you share code to reproduce this issue please?

[Hyper-Devil]

@scottire Sorry to bother you, this is my log.

2022-08-12 21:29:26.829 | INFO | yolox.core.trainer:before_train:136 - Model Summary: Params: 8.94M, Gflops: 26.76
2022-08-12 21:29:28.867 | INFO | yolox.core.trainer:before_train:155 - init prefetcher, this might take one minute or less...
2022-08-12 21:29:50.369 | ERROR | yolox.core.launch:_distributed_worker:147 - An error has been caught in function '_distributed_worker', process 'SpawnProcess-1' (78), thread 'MainThread' (140542829814976):
Traceback (most recent call last):

File "", line 1, in
File "/usr/local/lib/python3.9/multiprocessing/spawn.py", line 116, in spawn_main
exitcode = _main(fd, parent_sentinel)
│ │ └ 5
│ └ 8
└ <function _main at 0x7fd2ad313940>
File "/usr/local/lib/python3.9/multiprocessing/spawn.py", line 129, in _main
return self._bootstrap(parent_sentinel)
│ │ └ 5
│ └ <function BaseProcess._bootstrap at 0x7fd2ad41ca60>
└
File "/usr/local/lib/python3.9/multiprocessing/process.py", line 315, in _bootstrap
self.run()
│ └ <function BaseProcess.run at 0x7fd2ad41c0d0>
└
File "/usr/local/lib/python3.9/multiprocessing/process.py", line 108, in run
self._target(*self._args, **self._kwargs)
│ │ │ │ │ └ {}
│ │ │ │ └
│ │ │ └ (<function _distributed_worker at 0x7fd120eccee0>, 0, (<function main at 0x7fd11498f790>, 2, 2, 0, 'nccl', 'tcp://127.0.0.1:4...
│ │ └
│ └ <function _wrap at 0x7fd1212ad700>
└
File "/usr/local/lib/python3.9/site-packages/torch/multiprocessing/spawn.py", line 59, in _wrap
fn(i, *args)
│ │ └ (<function main at 0x7fd11498f790>, 2, 2, 0, 'nccl', 'tcp://127.0.0.1:40017', (╒═══════════════════╤═════════════════════════...
│ └ 0
└ <function _distributed_worker at 0x7fd120eccee0>

File "/root/YOLOX-docker/yolox/core/launch.py", line 147, in _distributed_worker
main_func(*args)
│ └ (╒═══════════════════╤═══════════════════════════════════════════════════════════════════════════════════════════════════════...
└ <function main at 0x7fd11498f790>

File "/root/YOLOX-docker/tools/train.py", line 118, in main
trainer.train()
│ └ <function Trainer.train at 0x7fd1143de4c0>
└ <yolox.core.trainer.Trainer object at 0x7fd1143f4b80>

File "/root/YOLOX-docker/yolox/core/trainer.py", line 74, in train
self.before_train()
│ └ <function Trainer.before_train at 0x7fd1143ded30>
└ <yolox.core.trainer.Trainer object at 0x7fd1143f4b80>

File "/root/YOLOX-docker/yolox/core/trainer.py", line 183, in before_train
self.wandb_logger = WandbLogger.initialize_wandb_logger(
│ │ └ <classmethod object at 0x7fd11f38ae50>
│ └ <class 'yolox.utils.logger.WandbLogger'>
└ <yolox.core.trainer.Trainer object at 0x7fd1143f4b80>

File "/root/YOLOX-docker/yolox/utils/logger.py", line 384, in initialize_wandb_logger
return cls(config=vars(exp), val_dataset=val_dataset, **wandb_params)
│ │ │ └ {'project': 'yolox-s'}
│ │ └ <yolox.data.datasets.voc.VOCDetection object at 0x7fd146cec5e0>
│ └ ╒═══════════════════╤════════════════════════════════════════════════════════════════════════════════════════════════════════...
└ <class 'yolox.utils.logger.WandbLogger'>

File "/root/YOLOX-docker/yolox/utils/logger.py", line 206, in init
self.cats = val_dataset.cats
│ └ <yolox.data.datasets.voc.VOCDetection object at 0x7fd146cec5e0>
└ <yolox.utils.logger.WandbLogger object at 0x7fd146f0ddc0>

AttributeError: 'VOCDetection' object has no attribute 'cats'

[manangoel99]

Hi @samithaFHSS ! The reason this error is coming up is because the logger only currently supports COCO datasets and not VOC datasets. We will try to add it ASAP