This “meta-integrator” repeatedly invokes a provided sub-integrator until the computed radiance values satisfy a specified relative error bound (5% by default) with a certain probability (95% by default).
Internally, it uses a Z-test to decide when to stop collecting samples. While repeatedly applying a Z-test in this manner is not good practice in terms of a rigorous statistical analysis, it provides a useful mathematically motivated stopping criterion.
- Max Relative Error : float
- Maximum relative error threshold (Default: 0.05)
- Required P-value : float
- Required p-value to accept a sample (Default: 0.05)
- Max Number of Samples : integer
- Maximum number of samples to be generated relative to the number of configured pixel samples. The adaptive integrator will stop after this many samples, regardless of whether or not the error criterion was satisfied. A negative value will be interpreted as ∞. (Default: 32—for instance, when 64 pixel samples are configured in the sampler, this means that the adaptive integrator will give up after 32*64=2048 samples)