Overview

SurveyAnalytica’s Image Tagging module provides a complete workflow for organizing, labeling, and preparing image datasets for machine learning. Whether you are building an image classification model, curating a product catalog, or analyzing visual content from surveys, the Image Tagging tools give you auto-tagging via AI-powered computer vision, manual tagging interfaces, bulk operations, and direct export to your analytics engine.

Getting Started with Image Tagging

Uploading Images

Upload images through the Data Management wizard. You can upload individual images, multiple images at once, or a ZIP archive containing images organized in folders. Supported image formats include PNG, JPG, JPEG, GIF, BMP, and WebP.

Indexing Images

After uploading, images need to be indexed. The indexing process scans your uploaded files and creates records in the image tagging database with full metadata. Navigate to your image dataset and the system will automatically attempt to index images from your upload folders.

During indexing, the system extracts:

• File metadata: File name, extension, size, content type, upload date
• Folder structure: Path segments, parent folders
• Classification hints: Automatically detects class names from folder names and train/test/validation split indicators from folder paths
• Suggested tags: Generated from folder-based class names and split assignments

Use Force re-index to delete all existing image records and re-index from scratch.

Auto-Tagging with Computer Vision

The auto-tagging feature uses AI-powered computer vision to automatically analyze images and generate descriptive tags.

How to Auto-Tag

1. Navigate to your image dataset.
2. Click Auto-Tag to start the process.
3. The system processes all images in batch, analyzing each image using computer vision.
4. Results are saved with tags, confidence scores, and the source marked as auto-generated.

Tag Structure

Each tag includes:

• tag: The label text (e.g., “dog”, “outdoor”, “mountain”)
• source: How the tag was created — auto-generated, manual, folder, classification, split
• confidence: A score from 0 to 1 indicating the certainty of the tag
• approved: Boolean indicating whether the tag has been reviewed and approved

Manual Tagging

You can manually add, edit, or remove tags from any image:

• Add tags: Type new tag names and assign them to individual images or in bulk.
• Edit tags: Modify existing tag labels or confidence scores.
• Remove tags: Delete incorrect or unwanted tags.
• Approve tags: Review auto-generated tags and mark them as approved.

Bulk Operations

For large datasets, bulk operations are essential:

Bulk Tag

Apply tags to multiple images at once. You can target images by:

• Explicit image IDs: Select specific images to tag.
• Filter criteria: Apply tags to all images matching a split, class name, or folder.
• Mode: Choose “append” to add tags without removing existing ones, or “replace” to overwrite all tags.

Bulk Approve

Review and approve tags across multiple images simultaneously. Approve all tags for selected images or for the entire dataset.

Promote Suggestions

Convert suggested tags (generated from folder structure, file names, or split detection) into confirmed tags. Select which suggestion sources to promote (folder, split, filename).

Bulk Restore

Restore tags for multiple images at once, useful for undo operations after bulk changes.

Filtering and Search

The image tagging interface provides powerful filtering:

• By split: Filter images by train, test, validation, or dev split
• By class name: Filter by detected or assigned class label
• By folder: Filter by source folder
• By search: Search by file name or image ID
• By tag status: View untagged, tagged, or approved images

Facets are computed dynamically with cascading behavior — selecting a split filter updates the available class names accordingly.

Sorting Options

• Filename (ascending/descending)
• Date (ascending/descending)
• Tag count (ascending/descending)

Image Exclusion

Exclude problematic images from your dataset without deleting them:

• Mark images as excluded with a reason (manual, duplicate, low quality, etc.)
• Restore excluded images when needed
• View excluded image counts broken down by reason
• Excluded images are automatically filtered out during training data generation

Image Statistics

The image stats feature provides detailed metadata about your dataset:

• Dimensions: Min, max, mean, and median width and height across all images
• Formats: Distribution of image formats (JPEG, PNG, etc.)
• File sizes: Min, max, mean file sizes in KB and total size in MB
• Outliers: Images with dimensions more than 3x the median, flagged for review

Classification Workflow

For building image classification models, the platform provides a complete classification pipeline:

Analyze Dataset Structure

The system scans your folder structure to detect classes, splits, and unclassified images. It supports both standard folder analysis and AI-powered smart analysis.

Label Assignment

Labels can be assigned from multiple sources:

• Folder structure: Automatically detected from subfolder names
• CSV file: Upload or select a CSV file with filename-to-label mappings. The system auto-detects filename and label columns.
• Existing tags: Convert approved image tags into classification labels.
• Manual assignment: Assign labels to specific images through the UI.

Class Imbalance Detection

The platform analyzes class distribution and reports:

• Per-class counts and percentages
• Imbalance ratio (largest class / smallest class)
• Severity level (low, moderate, high)
• Recommendations for balancing (undersample, oversample, or collect more data)

Dataset Balancing

Automatically balance your dataset by excluding excess images from larger classes (undersampling). The system uses random sampling with a fixed seed for reproducibility.

Training Data Generation

Generate train/validation/test splits with configurable ratios:

• Default split: 70% train, 15% validation, 15% test
• Stratified sampling: Ensures each class is proportionally represented in each split
• Export formats: Folder structure (images organized into train/val/test folders) and CSV manifest
• Exclude excluded images: Automatically filters out excluded images from training data

Exporting Data

Export approved tags and image metadata for analytics. The export creates a structured dataset with image ID, file path, file name, tags, and custom fields, enabling analysis of your image dataset.

Tag History

All tagging operations are recorded in a history log, including bulk operations, manual edits, promotions, approvals, and restores. This provides a complete audit trail of how your dataset was curated over time.