Appearance

Frequently Asked Questions (FAQ) ​

🚀 Getting Started ​

How do I start using Rosepetal? ​

  1. Log in with the credentials provided by your administrator
  2. Explore the dashboard to familiarize yourself with the interface
  3. Create your first dataset by uploading sample images
  4. Label some images to train your model
  5. Train your first model with the labeled data

What do I need to get started? ​

  • Platform access: URL and credentials
  • Sample images: Minimum 50-100 per class
  • Compatible browser: Chrome, Firefox, Safari or Edge
  • Internet connection: To upload data and train models

Can I try without real data? ​

Yes, you can use the sample images included in the platform to practice with labeling and training tools before using your own data.

📁 Dataset Management ​

How many images do I need to train? ​

  • Classification: Minimum 100 images per class, ideally 300-500
  • Object detection: 50-200 examples per object type
  • Segmentation: 20-100 images with precise masks
  • Anomaly detection: 200+ "normal" images

What image formats does it accept? ​

The platform supports the most common formats:

  • JPG/JPEG: Recommended for photographs
  • PNG: Ideal for images with transparencies
  • BMP: Uncompressed format
  • TIFF: For high-quality images

Is there a limit on image size? ​

  • Recommended size: 224x224 to 1024x1024 pixels
  • Maximum weight: 10MB per image
  • Minimum resolution: 64x64 pixels
  • Images are automatically resized during training

Can I delete images after uploading them? ​

Yes, you can delete individual images from the labeling view or delete complete datasets from the main view. Be careful, this action cannot be undone.

🏷️ Labeling and Annotation ​

How do I correct an incorrect label? ​

  1. Open the dataset and go to the "Labeling" tab
  2. Navigate to the image with incorrect label
  3. Select the appropriate tool (classification, bounding box, etc.)
  4. Modify the annotation - it saves automatically

Are annotations saved automatically? ​

Yes, all annotations are saved automatically while you work. You don't need to manually click "save".

Can I label in batches? ​

For classification, you can use bulk tag assignment by selecting multiple images and applying the same label to all.

What do I do if an image is ambiguous? ​

  • Establish clear criteria before starting labeling
  • Document edge cases in dataset comments
  • Consult with the team to maintain consistency
  • Consider creating a specific class for ambiguous cases

🤖 Model Training ​

How long does it take to train a model? ​

It depends on several factors:

  • Number of images: More data = more time
  • Model type: Classification (1-2h), Detection (4-8h), Segmentation (8-12h)
  • Available hardware: GPU significantly accelerates
  • Configuration: More epochs = more time

Can I cancel a training in progress? ​

Yes, you can cancel at any time from the interface. Progress will be lost but won't affect other processes.

What does "accuracy" mean and what value is good? ​

  • Accuracy: Percentage of correct predictions
  • Good: >90% for simple classification
  • Excellent: >95% for critical applications
  • Consider: That it depends on the specific problem

My model has low precision, what do I do? ​

  1. Verify data quality: Correct and consistent labels
  2. Increase dataset: More images per class
  3. Balance classes: Similar number of examples
  4. Review parameters: Adjust training configuration
  5. Data augmentation: Activate data augmentation

⚙️ Technical Issues ​

The page loads slowly ​

  1. Check internet connection
  2. Clear browser cache
  3. Close unnecessary tabs
  4. Restart browser
  5. Try in incognito window

I can't upload images ​

  • Check format: JPG, PNG, BMP, TIFF only
  • Check size: Maximum 10MB per image
  • Review connection: Stable internet required
  • Permissions: Confirm you have write access

Training constantly fails ​

  1. Review logs: Look for specific error messages
  2. Verify dataset: No corrupted images
  3. Check balance: Sufficient examples per class
  4. Contact support: If problem persists

I don't see some modules in the dashboard ​

This is normal - modules are shown according to your user role:

  • Administrator: All modules
  • Standard user: Datasets, models, workflows
  • Viewer: Read-only information

🔧 Configuration and Customization ​

Can I change the language? ​

Yes, the platform supports Spanish and English. You can change from user settings or using the buttons in the interface.

How do I activate dark mode? ​

Use the dark mode switcher at the top of the interface. The change is immediate and remembered for future sessions.

Can I customize the dashboard? ​

Currently the dashboard layout is fixed, but automatically adapts to your screen size and user role.

🔐 Security and Privacy ​

Is my data secure? ​

Yes, the platform implements:

  • Encryption: Data in transit and storage
  • Authentication: Firebase Authentication
  • Backups: Regular automatic backups
  • Access control: By user roles

Can I download my data? ​

Yes, you can export complete datasets with images and annotations from the interface, or download trained models for external use.

Who can see my datasets? ​

Only you and system administrators. Datasets are private by default, although administrators can configure shared datasets.

📊 Performance and Limits ​

Are there limits on the number of datasets? ​

Limits depend on your subscription plan:

  • Standard users: Generally 10-50 datasets
  • Administrators: No specific limit
  • Storage: Limited by available space

Can I train multiple models simultaneously? ​

Depends on system resources. Generally one training per user at a time is allowed to optimize resources.

What if my model is very slow? ​

  • Automatic optimization: The platform optimizes models for production
  • Hardware upgrade: Consider upgrading GPU for better performance
  • Model pruning: Techniques to reduce size without losing precision

🔄 Integrations and APIs ​

Can I integrate with other systems? ​

Yes, through:

  • REST APIs: For programmatic integration
  • Node-RED flows: For visual automation
  • Webhooks: For notifications
  • Exports: Data in standard formats

Is there an API for predictions? ​

Yes, each trained model exposes a REST API for making predictions. Documentation is available in the API section.

Can I use models outside the platform? ​

Models can be exported in standard formats (ONNX, TensorFlow, PyTorch) for use in other systems.

🆘 Support and Help ​

Where can I find more help? ​

  • Complete documentation: In this same section
  • Glossary: For technical terms
  • Troubleshooting: For common errors
  • Technical support: Contact your administrator

Can I request new features? ​

Yes, improvement suggestions are welcome. Contact the development team through your system administrator.

Is training available? ​

Check with your organization about available training sessions. Documentation includes step-by-step guides for self-learning.

How do I report a bug? ​

  1. Document the problem: Steps to reproduce
  2. Include screenshots: If possible
  3. Browser information: Version and operating system
  4. Contact support: Through established channels

🔍 Specific Use Cases ​

Can I detect multiple defects at once? ​

Yes, using object detection you can train a model to find different types of defects simultaneously in the same image.

Does it work with grayscale images? ​

Yes, the platform automatically handles color and grayscale images. Models adapt to the image type during training.

Can I measure defect areas? ​

With segmentation you get pixel-precise masks that allow calculating exact areas of defects or regions of interest.

How do I handle images of different sizes? ​

The platform automatically resizes images during processing, maintaining aspect ratio when possible.