Open ckLibra opened 9 months ago
the contraction function in current version may cause numerical instability in torch.pow(2x_mag_sqrt - 1, 1/3), when 2x_mag_sqrt - 1 < 0. According to the original paper (Eq. 14 in ZipNeRF), I revise it to ensure 2*x_mag_sqrt - 1 >= 1.
the contraction function in current version may cause numerical instability in torch.pow(2x_mag_sqrt - 1, 1/3), when 2x_mag_sqrt - 1 < 0. According to the original paper (Eq. 14 in ZipNeRF), I revise it to ensure 2*x_mag_sqrt - 1 >= 1.