SynthIdImageOptions
Options & Configuration · AiDotNet.Safety.Watermarking
Configuration of a SynthIdImageWatermarker. It adds SynthID-Image’s detection objective and
calibrated decision to the StegaStamp networks and distortions configured by StegaStampOptions.
Inheritance: ModelOptions → NeuralNetworkOptions → StegaStampOptions → SynthIdImageOptions
How It Works
The defaults follow what Gowal et al. (2025, Sec. 8) state for SynthID-O: a 136-bit payload in 512 x 512 images,
with thresholds set for a 0.1% false positive rate. The paper does not describe error correction, so all 136 bits are
payload here. It does not publish its networks, losses or training schedule. Those come from
StegaStampOptions.
Constructors
| Constructor | Summary |
|---|---|
SynthIdImageOptions() |
SynthID-O’s stated sizes, 512 x 512 images and 136 payload bits, with no error correction. |
SynthIdImageOptions(SynthIdImageOptions other) |
A copy of other. |
Properties
| Property | Summary |
|---|---|
double DetectionLossScale { get; set; } |
Weight of the detection loss: cross-entropy of the detection logit on distorted watermarked images (label 1) and on the same distortions of the clean images (label 0). |
double FalsePositiveRate { get; set; } |
The level alpha of the calibrated decision. |
double SpecificityLossScale { get; set; } |
Weight of the specificity loss. |