Collect indexes of duplicates values in 2D int array algorithm

I'm working on slot machine and faced the problem of collecting outcomes results. Question is what is the fastest approach to collect indexes of duplicate values in 2D int array? Condition here is to collect only idexes of values which occur 5 times

CASE 1

input (get indexes of 3 value only):

int[][] input = new int[][]{
            new int[]{1, 2, 3, 4, 8},
            new int[]{6, 3, 2, 3, 5},
            new int[]{3, 9, 7, 1, 3}
    };

expected output:

[2, 1, 0, 1, 2]

CASE 2

input (get indexes of 3 and 5 values only):

int[][] input = new int[][]{
            new int[]{1, 5, 3, 5, 8},
            new int[]{5, 3, 5, 3, 5},
            new int[]{3, 9, 7, 1, 3}
    };

expected output:

[2, 1, 0, 1, 2] //for 3 value
[1, 0, 1, 0, 1] //for 5 value

MY SOLUTION(its quite poor)

1) gather duplicates (this one not working for CASE 2)

Map<Integer, Integer> amountMap = new HashMap<>();
for (int[] row : railSpin) {
    for (int value : row) {
        amountMap.put(value, amountMap.containsKey(value) ? amountMap.get(value) + 1 : 1);
    }
}

2) remove non 5 matches

if (amountMap.containsValue(5)) {
    Iterator<Integer> amountIterator = amountMap.values().iterator();
    while (amountIterator.hasNext()) {
        if (amountIterator.next() != 5) {
            amountIterator.remove();
        }
    }
}

3) iterate upside-down and collect indexes

List<Integer> indexes = new ArrayList<>();
for (int row = 0; row < 5; row++) {
    for (int col = 0; col < railSpin.length; col++) {
        int valueToCheck = railSpin[col][row];
        if (amountMap.keySet().contains(valueToCheck)) {
            indexes.add(col);
        }
    }
}

4) split array if needed

List<List<Integer>> splitResults = new ArrayList<>();
for (int start = 0; start < indexes.size(); start += 5) {
    int end = Math.min(start + 5, indexes.size());
    List<Integer> sublist = indexes.subList(start, end);
    splitResults.add(new ArrayList<>());
    splitResults.get(start /5).addAll(sublist);
}

Can you suggest a solution without so many iterations and which would be suitable for CASE 2? I belive in power of stackoverflow