I have this tensor dimension:

(batch_size, class_id, range_indices) -> (4, 3, 2)
int64
[[[1250 1302]
  [1324 1374]
  [1458 1572]]

 [[1911 1955]
  [1979 2028]
  [2120 2224]]

 [[2546 2599]
  [2624 2668]
  [2765 2871]]

 [[3223 3270]
  [3286 3347]
  [3434 3539]]]

How do I construct densed representation with filled value with this rule:

Since there is 3 class IDs, therefore:

1. Class ID 0: filled with 1
2. Class ID 1: filled with 2
3. Class ID 2: filled with 3
4. Default: filled with 0

Therefore, it will outputting vector like this:

[0 0 0 ...(until 1250)... 1 1 1 ...(until 1302)... 0 0 0 ...(until 1324)... 2 2 2 ...(until 1374)... and so on]

Here is copiable code:

data = np.array([[[1250, 1302],
                  [1324, 1374],
                  [1458, 1572]],

                 [[1911, 1955],
                  [1979, 2028],
                  [2120, 2224]],

                 [[2546, 2599],
                  [2624, 2668],
                  [2765, 2871]],

                 [[3223, 3270],
                  [3286, 3347],
                  [3434, 3539]]])