images resolution 2376 × 1584 #14335
Replies: 2 comments 3 replies
-
Hi @ovidedecroly, Thank you for reaching out! Using high-resolution images like 2376×1584 with YOLOv8 on an A100 GPU is definitely feasible, but there are a few considerations to keep in mind to optimize performance and manage resources effectively. Key Considerations and Recommendations:
Example Configuration:Here's a sample configuration snippet for training with high-resolution images: from ultralytics import YOLO
# Load a model
model = YOLO('yolov8n.pt') # or your custom model
# Train the model
results = model.train(
data='your_dataset.yaml',
epochs=100,
imgsz=(2376, 1584),
batch=4, # Adjust based on your GPU memory
amp=True, # Enable mixed precision training
cache='ram' # Cache images in RAM
) Additional Resources:For more detailed guidance, you can refer to our documentation on training and minimum reproducible example to ensure you have all the necessary steps covered. Feel free to reach out if you have any more questions. Happy training! 🚀 |
Beta Was this translation helpful? Give feedback.
-
Hi @glenn-jocher I have two primary sources of data: a public dataset and my own private dataset. I am reaching out to seek your expertise on the best practices for combining these datasets effectively for training the YOLOv8 model. Here are some details about my datasets: Public Dataset: Number of image public dataset per species: number of image private (own) dataset: Given these variations, I am considering several techniques but would appreciate your insights on the following: Transfer Learning with Fine-Tuning: Using the public dataset for initial training and then fine-tuning with the private dataset. Does this approach make sense given the difference in resolution and quantity of images? Domain Adaptation: If the distribution between the public and private datasets differs significantly, would you recommend domain adaptation techniques? If so, which methods would be most effective? Data Splitting Strategy: How should I approach splitting the combined dataset for training and validation? Should I augment all images, or focus augmentation on the underrepresented classes? Weighted Loss Function: Given the imbalance in the number of images per species, would applying a weighted loss function be beneficial to ensure the model does not favor the more represented classes? Any Other Recommendations: Are there any other techniques or best practices you would suggest for effectively combining and training on these datasets? I would greatly appreciate any guidance or suggestions you can provide. Thank you for your time and support. Best regard |
Beta Was this translation helpful? Give feedback.
-
Hi @glenn-jocher
I hope you are doing well. I have a question regarding the usage of high-resolution images with YOLOv8. Specifically, I am planning to use images with a resolution of 2376×1584 for training, validation, and testing purposes. I am using an A100 GPU for this task. I would like to know if YOLOv8 can effectively handle this resolution for these tasks.
Could you please provide insights or recommendations on the following points:
Are there any limitations or considerations I should be aware of when using YOLOv8 with such high-resolution images?
What impact might this resolution have on the training time and GPU memory usage?
Are there any specific settings or configurations in YOLOv8 that I should adjust to optimize performance with high-resolution images?
Would you recommend any particular strategies to manage potential challenges associated with high-resolution image processing in YOLOv8?
Thank you for your assistance.
Best regards,
Beta Was this translation helpful? Give feedback.
All reactions