Find the useless pixels
Some pixels never change across all 1,797 samples — they carry zero information. Drag the slider to set a minimum variance threshold and watch useless pixels disappear.
Pixel variance (std dev)
Useful pixels
Loading...
Loading...
Loading...
Think Deeper
Try this:
In the Explore tab, drag the variance slider to max. How many pixels survive? Could a model still work with just those?
At high thresholds, only ~10-15 center pixels remain — and yes, a model can still classify reasonably well. This is dimensionality reduction: fewer features can mean faster training and less overfitting.
Cybersecurity tie-in: In security logs, this is equivalent to a field that's always
empty — like a TTL column that's always 64, or a protocol field that's always TCP. Removing useless
features makes models faster and often more accurate.