Operation Process
This example demonstrates the process of generating synthetic data.
This document outlines the process of generating synthetic images by introducing defects (e.g., break) into scenes (e.g., chestnuts). The workflow includes image uploading, annotation, scene synthesis, and model validation.
1.Create Model
1. Click "Create a model".
2. Enter the Model Name: Provide a unique and descriptive name.
3. Select the Generation Model:
￮ Fixed Position Generator: Generates features seamlessly within the specified labeled range.
￮ Random Position Generator: Randomly generates one or more features within the labeled range.
4. Click "Create" to create the model.
2.Upload Reference Images
1. Click "+img" to upload reference images showcasing the features to be learned.
2. Ensure the images meet the upload requirements, then click "Upload now" to proceed.
3. Batch uploading is supported. Once completed, verify the uploaded images.
For locally labeled images, refer to +img and labels
3.Label Features
1. Click "+ Add a label type" to create a new label category.
2. Use the annotation tools to label features:
￮ Brush Tool: For free-form annotations, ideal for irregular shapes.
￮ Polygon Tool: For precise annotations by connecting points.
￮ Drag Tool: To move or adjust annotated areas.
￮ Copy Tool: To duplicate annotations across multiple images.
3. Click "Add Background" to proceed.
4.Upload Background Images
1. Click "+img" to upload background images.
2. Ensure the images meet the upload requirements, then click "Upload now".
5.Label Synthetic Areas
1. Use the annotation tools to label areas for synthesis on the background image.
2. Click "Preview" to continue.
6.Preview
1. Select a Label for preview from the dropdown menu.
2. Click "Preview" to generate a synthesized image.
3. Adjust composition parameters if needed:
￮ Control Strength: Higher values retain more reference image details.
￮ Guidance Scale: Higher values increase fidelity to the reference image.
￮ Mixedness: Higher values blur the reference image.
￮ Shape Control: Optimizes annotation shape consistency.
￮ Disable Rotation: Maintains original orientation.
4. Click "Default" to reset parameters.
7.Curate
1. Set the desired quantity for each label using the slider.
2. Click "Generate" to initiate image synthesis.
8.Generate
Click "Start Generating" to begin the image synthesis process.
9.Synthetic
1. Single Image Selection: Click the star icon to add an image to your favorites.
2. Bulk Operations:
￮ Add to Favorites: Save selected images.
￮ Remove from Favorites: Remove selected images.
￮ Delete: Permanently delete selected images.
3. Click "Start training" to proceed.
10.Train
1. Select synthetic images, reference images, or both for training.
2. Click "Start training" to begin model training.
11.Test images
1. Click "+img" to upload test images.
2. Alternatively, click "Add Reference" to use reference images as test images.
12.Test
Click "Test" to start the test. After the test is completed, click "View the test result" to check the test result.
13.Download
1. Images: Download synthesized images (free quota: 100 images/month).
2. Model: Choose hardware and model version for export.
3. SDK: Select the SDK version and download it.
Still have a question?
Each section should be concise, user-friendly, and direct users to additional resources or documentation when necessary.