Dataset Statistics

The Dataset Statistics page helps you explore training data distribution across tags, view image counts, and analyze class breakdown information. Use it to understand your dataset composition and identify potential imbalances.

Page Overview

The page displays:

• Statistics cards - Total tags, total models, best mAP50, and total classes
• Tag filter - Filter all data by a specific tag
• Class Breakdown - Detailed class-level statistics with search
• Tags table - Expandable rows showing tag-level and model-level data

Class Breakdown

The Class Breakdown section shows detailed statistics for each class in your dataset.

Features:

• Search - Filter classes by name for quick lookup
• Summary stats - Total classes, train images, and test images
• Pagination - Navigate through large class lists

Column	Description
Class Name	Name of the detection class
Train Images	Number of training images containing this class
Train BBoxes	Total bounding box annotations for training
Test Images	Number of test images containing this class
Test BBoxes	Total bounding box annotations for testing

Tip: Use the tag filter at the top to see class breakdown for a specific tag only.

Tags Table

The Tags table shows aggregate statistics for each tag. Columns are sortable by clicking the header.

Column	Description
Tag	Tag name (click row to expand)
Models	Number of trained models for this tag
Train Images	Total training images across all models
Test Images	Total test images across all models
Best mAP50	Highest accuracy achieved by any model for this tag

Viewing Tag Models

Click any row to expand it and see all models for that tag.

The expanded view shows:

• Model list - All models for the tag with individual statistics
• Average mAP50 - Average accuracy across all models
• Pagination - Navigate through models if there are many

Model Class Breakdown

Click any model row in the expanded view to see its per-class breakdown.

This dialog shows how training and test data is distributed across classes for a specific model, helping identify class imbalances.

Filtering Data

Use the Filter by Tag dropdown at the top of the page to focus on a specific tag. When filtered:

• Stats cards update to show filtered totals
• Class Breakdown shows only classes for that tag
• Tags table shows only the selected tag

Select "All Tags" to clear the filter and see complete data.

Sorting Data

Click any column header in the Tags table to sort by that column. Click again to toggle between ascending and descending order.

Sortable columns:

• Tag (alphabetical)
• Models (by count)
• Train Images (by count)
• Test Images (by count)
• Best mAP50 (by score)

Tips for Data Analysis

• Compare Train vs Test image counts to ensure balanced splits
• Check for classes with very few images that might need more data
• Use the search in Class Breakdown to quickly find specific classes
• Sort by Best mAP50 to see which tags are performing well
• Expand tags to see model progression over time