Building Reliable ColorBase Datasets from Swatch Scans

Organise swatch cards before scanning

• Group markers by brand or collection so metadata stays tidy in the resulting dataset.
• Label the physical sheets with the same identifiers you plan to use in ColorSpace.
• Inspect for ink pooling, glare, or stray marks that could throw off segmentation.

Capture at a consistent resolution

• ColorBase supports TIFF, PNG, and JPG up to 50 MB per file as referenced on the ColorBase Builder page.
• Use 300–600 DPI depending on the grain of your paper stock; higher DPI helps the pipeline detect subtle gradients.
• Disable auto-enhancement modes on scanners or cameras to keep LAB readings true to the original pigment.

Calibrate lighting and white balance

• Place a neutral grey reference patch beside the swatches, especially when working with camera captures.
• Match the capture lighting to the environment where markers are typically used to minimise adaptation surprises.
• Avoid mixed colour temperature bulbs; daylight-balanced LEDs or indirect natural light produce the most consistent results.

Upload and segmentation

Once a scan is ready, the ColorBase pipeline parses individual chips, skipping blank gaps and labelling rows automatically. This mirrors the description on the ColorBase Builder page where multi-swatch parsing removes manual cropping.

If the dataset is destined for classroom demonstrations, keep the cards aligned during capture; straight edges give the segmentation model more reliable anchor points.

Noise reduction and LAB calibration

ColorBase applies adaptive smoothing to reduce scanner noise while guarding delicate gradients—especially important for broad chisel strokes that show subtle streaks.

The pipeline stores both RGB and LAB values for every entry, allowing ColorSpace to run harmony math in LAB while still presenting colours using the measured RGB values in the UI.

Metadata confirmation and versioning

Before exporting, you can verify marker codes, names, brand attribution, and any notes that should accompany the dataset. Keeping metadata consistent ensures the Marker Collection view remains searchable.

Each upload is versioned, giving teams the option to roll back if a later scan set introduces issues. Version labels can mirror product release cycles or classroom semesters.