Is there a way to hide/suppress any console output when loading and inferencing via PyTorch Hub?

[haimat]

❔Question

I would like to use YOLOv5 in a custom project, so I need to be in control of what will be printed to the console. Currently I am loading the model and using it like described in your wiki:

model = torch.hub.load("ultralytics/yolov5", "custom", path_or_model=args.model)
model.conf = args.confidence
model.iou = args.iou
img = Image.open(args,image)
results = model(img, size=args.size, augment=True)
preds = results.pandas().xyxy[0]

However, these line result in some console output (like network layout, autoShape info, etc.)
Is there any way to hide all these standad console output?

[github-actions]

👋 Hello @haimat, thank you for your interest in 🚀 YOLOv5! Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution.

If this is a 🐛 Bug Report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset images, training logs, screenshots, and a public link to online W&B logging if available.

For business inquiries or professional support requests please visit https://www.ultralytics.com or email Glenn Jocher at glenn.jocher@ultralytics.com.

Requirements

Python 3.8 or later with all requirements.txt dependencies installed, including torch>=1.7. To install run:

$ pip install -r requirements.txt

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Google Colab and Kaggle notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training (train.py), testing (test.py), inference (detect.py) and export (export.py) on MacOS, Windows, and Ubuntu every 24 hours and on every commit.

[glenn-jocher]

@haimat yes we probably want to group these outputs into a verbose=True setting. If you'd like to update the PyTorch Hub model code for this and submit a PR that would be awesome! If not we can add this to our TODO queue, which is very long at the moment.

[glenn-jocher]

@haimat see https://github.com/ultralytics/yolov5/blob/master/hubconf.py for the right place to start.

yolov5/hubconf.py

Lines 1 to 7 in 1df8c6c

	"""YOLOv5 PyTorch Hub models https://pytorch.org/hub/ultralytics_yolov5/
	Usage:
	import torch
	model = torch.hub.load('ultralytics/yolov5', 'yolov5s')
	"""

[haimat]

@haimat yes we probably want to group these outputs into a verbose=True setting. If you'd like to update the PyTorch Hub model code for this and submit a PR that would be awesome! If not we can add this to our TODO queue, which is very long at the moment.

@glenn-jocher Well, for now I always worked with TF and Keras, so this is my first project with PyTorch and its hub. This YOLOv5 project is pretty cool though and I am always willing to learn something new, so why not, I can give this change a try. But could you please be so kind and give me just a few hints where to start and how you would like to have this change? I am totally fine with Python in general and these things, I would just like to know what you would have in mind, so that my change can reflect that. Thanks in advance!

[glenn-jocher]

@haimat well, there's two ways of managing the output, either with if verbose: logger.info() statements or maybe by setting the logging level directly before the logging events. The second is preferred if possible, though I've never implemented this myself so I'm not sure if it makes sense.

The first is pretty simple, but will require a lot of code changes, plus passing the verbose value through several functions to all of the places that currently print/log output, which would be a hassle. The main place where the model architecture is printed as it's built is here:

yolov5/models/yolo.py

Lines 201 to 202 in 1df8c6c

	def parse_model(d, ch): # model_dict, input_channels(3)
	logger.info('\n%3s%18s%3s%10s %-40s%-30s' % ('', 'from', 'n', 'params', 'module', 'arguments'))

[haimat]

@glenn-jocher Ok thanks, I will have a look at thia. One more thing though: When I update the logging logic e.g. for the detect.py file and related modules, then this would not directly reflect the hub model, right? How can I then check whether my change would also work for the hub version of YOLOv5?

[glenn-jocher]

@haimat ah well yes once we merge a PR, new operations with the same Hub cache would be unaffected unless the Hub cache is updated with force_reload=True. You can think of force_reload as a git pull for pytorch hub.

But also detect.py is mostly a separate inference channel from pytorch hub.

[NanoCode012]

Hi, speaking of the first method that Glenn mentioned, maybe we can rework the logging logic a bit? It was left untouched for a while.

yolov5/utils/general.py

Lines 33 to 36 in 1df8c6c

	def set_logging(rank=-1):
	logging.basicConfig(
	format="%(message)s",
	level=logging.INFO if rank in [-1, 0] else logging.WARN)

Maybe we can a verbose option here? It could also be used with the train.py and test.py later as well by passing into this function.

def set_logging(rank=-1, verbose=True): 
     logging.basicConfig( 
         format="%(message)s", 
         level=logging.INFO if (verbose and rank in [-1, 0]) else logging.WARN) 

yolov5/detect.py

Lines 28 to 29 in 1df8c6c

	# Initialize
	set_logging()

The only thing to note is that if verbose is false, it should disable the majority/all the output (as far as I know).

[glenn-jocher]

@NanoCode012 yeah I think this is a great idea. I wonder if there's a way to pair this with a global verbose variable or an environment variable somehow so any function could set verbose=False and the other functions/files would be faithful to the change.

[haimat]

I was thinking of utilizing Python's logging functionalities. How about passing the log level (like "error", "warning", "info") to the script, and based on that more or less logging will be printed. And all that comes for free if we replace all print statements with the logger call.

[glenn-jocher]

@haimat @NanoCode012 yes these are both good ideas. I've added a TODO label here to remind ourselves to get to this eventually, but I don't have much availability for this and the priority is not high, so it may take a while. If either of you want to take a stab at it feel free to submit a PR, thank you!