clearml variables as environments

[mbenami]

Search before asking

• I have searched the YOLOv5 issues and found no similar feature requests.

Description

when you lunch yolo training with clearml you cannot control the clearml task variable from the train.py hyp.opt
like project_name task_name ..

yolov5/utils/loggers/clearml/clearml_utils.py

Line 87 in 82a5585

self.task = Task.init(

it will be nice if you could change them in the training input so you can lunch several training each with different names.

Use case

lunch different training with different clearml project_name or task_name without editing the code

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[glenn-jocher]

@mbenami that's a good suggestion!

@thepycoder is there a way for ClearML to use the --project and --name arguments for YOLOv5 training, i.e. if I do

python train.py --project PROJECT --name NAME

Then it would be nice to see this in ClearML as i.e. experiments/PROJECT/NAME. Currently all trainings go into a YOLOv5/ project.

[thepycoder]

Hi @glenn-jocher

Yeah your suggestion would work :)

I didn't include it in v1 integration as to not adjust the global parameters too much. But with your blessing I'll make a PR for it as soon as I'm back at a computer!

EDIT: somehow I missed it that the project and name variables already exist! The PR will be small and an easy fix :)

Thanks @mbenami for reporting!

[glenn-jocher]

@thepycoder awesome, that would be great, thanks :)

[mbenami]

Hi @thepycoder
is there an option also to add tags?
I thought to change

yolov5/utils/loggers/clearml/clearml_utils.py

Line 90 in 454dae1

tags=['YOLOv5'],

tags=getattr(opt, 'tags', ['YOLOv5'])
If we don't want to add input args
(you can use it with run(kwargs**)

[thepycoder]

Hi @mbenami

Sound like a good plan to me! We'd have to document it somewhere, so people know they can use it, but apart from that I see no reason why not!

@glenn-jocher Do you think it makes sense to just use kwargs, or is this a large enough feature to warrant its own argument here?

[glenn-jocher]

@AyushExel what do you think about adding a tags argument? Is this available in other loggers?