Image Metrics

Compute per-image quality metrics and add them as columns to your table. Cross-plot metrics in the Data Workbench to find outliers — dark, blurry, or noisy images that degrade training. Especially powerful after a training run: correlate image quality with per-sample loss, false positives, and embeddings to understand why your model struggles on certain images.

✨
15 Metrics
Brightness, sharpness, noise, contrast, entropy & more
🔗
Full Lineage
Results stored as EditedTable columns preserving table history
⚡
Parallel I/O
Multi-threaded image loading for fast processing
📊
Cross-plot
Correlate quality with training loss, FPs, and embeddings
1 Select tables
2 Choose metrics
3 Compute
Config

1 Select Tables

No tables selected. Use the action button on a table, or click + Add Table above.
Add one or more tables to compute metrics on. You can also launch this from the action button on any table page.

2 Choose Metrics

0 selected
Click Recommended for a quick start, or select individual metrics. Results are stored as new columns on an EditedTable.

3 Output

The suffix is appended to each input table name to create the output table (e.g. "train" becomes "train-metrics").