Question about grid_subsampling code

Thanks for your great work and nice code.

My question is about the code in grid_subsampling when computing the subsampled_class for the subsampled point. The code is:

https://github.com/HuguesTHOMAS/KPConv/blob/935a2bd75775b07eda8cecbacf2d91b2dd7cbf8d/cpp_wrappers/cpp_subsampling/grid_subsampling/grid_subsampling.cpp#L100

The logic here is to find out the label with the most occurrence number. The label’s occurrence number is stored in an unordered_map, with different labels as different keys, and corresponding occurrence number as values.

The code is using max_element to do it, but does not pass the third comp param. So what the code actually does, is just to find the max key in the unordered_map. It’s not what we want.

I think we need to add the third param to max_element like this:

max_element(v.second.labels[i].begin(), v.second.labels[i].end(), 
    [](const pair<int, int>&a, const pair<int, int>&b){return a.second < b.second;})

My test code is as follows:

#include <iostream>
#include <unordered_map>
#include <algorithm>

using namespace std;

int main() {
    unordered_map<int, int> map; // label -> occurrence number
    map[5] = 2;
    map[2] = 3;

    auto r1 = max_element(map.begin(), map.end());
    cout << r1->first << "->" << r1->second << endl; // output: 5->2, it's wrong!


    auto r2 = max_element(map.begin(), map.end(),
            [](const pair<int, int>&a, const pair<int, int>&b){return a.second < b.second;});
    cout << r2->first << "->" << r2->second << endl; // output: 2->3, it's right

    return 0;
}

[HuguesTHOMAS]

Thanks for the info

[LiyaoTang]

I see, probably the pointXYZ part let me think it’s from PCL. Particularly, I encountered this piece of code in RandLA-Net (here: https://github.com/QingyongHu/RandLA-Net/tree/master/utils/cpp_wrappers)