A Python implementation of the method: gridhopping.py. The PyTorch framework is used for vectorized computation. Approximately similar results could be achieved with NumPy.
You can use the script test.py to see how the method compares to the usual O(N^3) polygonization scheme:
python3 test.py 512
Note that the C implementation is faster for simpler models such as a sphere in test.py. However, the difference becomes large in favor of the vectorized version once we start dealing with SDFs represented by NNs or Fourier features.
The code for a more detailed experimental analysis is available here.