closing mosaic dataloader after the last 10 epochs about ultralytics

[github-actions]

👋 Hello @IDLEGLANCE, thank you for your interest in Ultralytics YOLOv8 🚀! We recommend a visit to the Docs for new users where you can find many Python and CLI usage examples and where many of the most common questions may already be answered.

If this is a 🐛 Bug Report, please provide a minimum reproducible example to help us debug it.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results.

Join the vibrant Ultralytics Discord 🎧 community for real-time conversations and collaborations. This platform offers a perfect space to inquire, showcase your work, and connect with fellow Ultralytics users.

Install

Pip install the ultralytics package including all requirements in a Python>=3.8 environment with PyTorch>=1.8.

pip install ultralytics

Environments

YOLOv8 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• 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 Ultralytics CI tests are currently passing. CI tests verify correct operation of all YOLOv8 Modes and Tasks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

[glenn-jocher]

@IDLEGLANCE hey there! It sounds like you're encountering a memory error related to the mosaic dataloader as it closes towards the last epochs of training. This can happen if the GPU runs out of memory when handling the combined data augmentation processes.

Here are a couple of suggestions to try:

1. Reduce the Image Size: Lowering the imgsz parameter can significantly reduce memory usage. Try setting it to 320 or 480 and see if that helps.

2. Adjust the Mosaic Loader: You can disable the mosaic data augmentation towards the end of training by setting close_mosaic=10 in your training YAML file. This might help in stabilizing the training as it concludes.

3. Monitor GPU Usage: Keep an eye on your GPU memory usage during training. Tools like nvidia-smi can be helpful to check if you're close to maxing out your GPU memory.

Here's an example command with a reduced image size:

yolo task=detect mode=train imgsz=480 data=test.yaml epochs=20 batch=5 name=test device=0

Try these adjustments and let us know how it goes! 😊

[IDLEGLANCE]

@IDLEGLANCE hey there! It sounds like you're encountering a memory error related to the mosaic dataloader as it closes towards the last epochs of training. This can happen if the GPU runs out of memory when handling the combined data augmentation processes.

Here are a couple of suggestions to try:

1. Reduce the Image Size: Lowering the imgsz parameter can significantly reduce memory usage. Try setting it to 320 or 480 and see if that helps.
2. Adjust the Mosaic Loader: You can disable the mosaic data augmentation towards the end of training by setting close_mosaic=10 in your training YAML file. This might help in stabilizing the training as it concludes.
3. Monitor GPU Usage: Keep an eye on your GPU memory usage during training. Tools like nvidia-smi can be helpful to check if you're close to maxing out your GPU memory.

Here's an example command with a reduced image size:

yolo task=detect mode=train imgsz=480 data=test.yaml epochs=20 batch=5 name=test device=0

Try these adjustments and let us know how it goes! 😊

hello, thanks for helping me but i already found out about the mosaic thingy so it doesnt stop anymore after the last 10 epochs. i just tried to put close_mosaic after the device=0 argument and then read the error code it gave me and it just said that there is a "=" missing thats how i found out.
the only thing i dont understand now is that im using 5 batch size and my gpu memory is at 1.4gb from 8.1gb and when im trying to increase the batch size it always says that theres not much memory left even tho its only at 1.4gb on 5 batch size. but its not a big deal since 1 epoch takes about 7-12 seconds.

[glenn-jocher]

Hey @IDLEGLANCE, glad to hear you resolved the mosaic issue! Regarding the GPU memory, it's possible that the memory reported as free (1.4GB) doesn't account for the peak usage during training, which can momentarily spike and cause out-of-memory errors when increasing the batch size. This is often due to how memory is allocated and managed within the training process. Keeping the batch size at a level that consistently works is a good strategy, especially with the quick epoch times you're experiencing. If you need to scale up, consider monitoring the memory usage more dynamically to catch those spikes. Happy training! 😊

[IDLEGLANCE]

Hey @IDLEGLANCE, glad to hear you resolved the mosaic issue! Regarding the GPU memory, it's possible that the memory reported as free (1.4GB) doesn't account for the peak usage during training, which can momentarily spike and cause out-of-memory errors when increasing the batch size. This is often due to how memory is allocated and managed within the training process. Keeping the batch size at a level that consistently works is a good strategy, especially with the quick epoch times you're experiencing. If you need to scale up, consider monitoring the memory usage more dynamically to catch those spikes. Happy training! 😊

thank you!

[glenn-jocher]

You're welcome! If you have any more questions or run into issues, feel free to reach out. Happy training! 😊