gaussian_kernel()

const std::vector< float > & MayaFlux::Kinesis::Vision::VisionExecutor::gaussian_kernel ( float  sigma )

private

Return a reference to the precomputed 1D Gaussian kernel for sigma.

Computes and caches on first call for a given sigma. The kernel is normalised to unit sum and has length 2*ceil(3*sigma)+1.

Definition at line 57 of file VisionExecutor.cpp.

{
    const auto key = std::bit_cast<uint32_t>(sigma);
    auto it = m_kernel_cache.find(key);
    if (it != m_kernel_cache.end())
        return it->second;
 
    const auto radius = static_cast<int32_t>(std::ceil(3.0F * sigma));
    const int32_t size = 2 * radius + 1;
    const float inv_2s2 = 1.0F / (2.0F * sigma * sigma);
 
    std::vector<float> k(static_cast<size_t>(size));
    float sum = 0.0F;
    for (int32_t i = -radius; i <= radius; ++i) {
        const float v = std::exp(-static_cast<float>(i * i) * inv_2s2);
        k[static_cast<size_t>(i + radius)] = v;
        sum += v;
    }
 
    auto kmap = Eigen::Map<Eigen::ArrayXf>(k.data(), static_cast<Eigen::Index>(k.size()));
    kmap /= sum;
 
    return m_kernel_cache.emplace(key, std::move(k)).first->second;
}

References k, m_kernel_cache, radius, and sigma.

Referenced by run().

Here is the caller graph for this function: