Are Potree Converter outputs deterministic or random?

[arunkothari84]

Hi, I’m using Potree 1.7 for Windows.
For my task, I need to edit a point cloud over the network — for example, delete or add other point clouds — and then serve the updated data from the server.

Because my point cloud is very large, I’m thinking of serving only the changed child nodes instead of the entire point cloud.

To make this possible, I need to confirm whether Potree Converter produces the same output each time when converting the same input point cloud.

I read in the Potree Converter documentation Link that the points are chosen randomly, but I wrote a small script (with ChatGPT’s help) to compare .bin files, and my findings were:

• When I converted the same point cloud twice, the results were identical.
• When I edited the point cloud slightly and re-converted, some .bin files were different, as expected.

So, my questions are:

1. Does the Potree Converter generate outputs deterministically (same result each time for the same input)?
2. If it uses randomness, is there a way to make it deterministic — for example, by setting a random seed or using specific converter options?

Thanks in advance for the clarification!

[christianmair]

Hi, to my understanding, PotreeConverter uses poisson-disk sampling by default (README), which would create reproducible results. This publication goes into more detail on how it works.

The random sampling strategy uses std::shuffle with a time-based seed. To make results reproducible, you'd have to overwrite the seed to a constant value. But I don't see a way to do that (other than compiling from source).

PotreeConverter/Converter/include/sampler_random.h

Lines 102 to 104 in c2328c4

	unsigned seed = std::chrono::system_clock::now().time_since_epoch().count();
	shuffle(indices.begin(), indices.end(), std::default_random_engine(seed));