vlarsson/lambdatwist

Re-implementation of Lambda Twist P3P solver (ECCV 2018)

lambdatwist's Issues

updated special case, and roots

Hi,
So the reference implementation was missing a special case, (which while extremely rare in real data happens to be a common unit test) which has been fixed. This further improved numerics as well and does so at no cost in terms of performance.

With regards to using the formula for the third order root, it makes a slight difference to numerics and performance, in general. But there is a special case which your code does not account for which is important, the root selected should be the one highest gradient magnitude , as if you have a double root it must be avoided. The root should also be non-zero.