TypeError: tuple indices must be integers or slices, not tuple

[yifanzhang666777]

if(d not in matched_indices[:,0]): TypeError: tuple indices must be integers or slices, not tuple

[eplatero97]

This happened to me when I changed:

• from sklearn.utils.linear_assignment_ import linear_assignment to
• from scipy.optimize import linear_sum_assignment as linear_assignment

Because sklearn's linear_assignment has been removed starting with version "0.23.0" and higher, I opted for Scipy's implementation of the function (which returns a tuple of np vectors as opposed to a matrix) (click here to learn more).

Thus, to get the equivalent result as if we were using sklearn's implementation I modified sort.py with below code:

import sklearn
skver = int(sklearn.__version__[2:4]) # extract middle number of version with format \d.\d\d.\d\d
if skver >= 23:
    from scipy.optimize import linear_sum_assignment as linear_assignment
    sk = False # we will not use sklearn implementation
else:
    from sklearn.utils.linear_assignment_ import linear_assignment
    sk = True # we will use sklearn implementation

# Below code was written after the `matched_indices` object had been created under the `associate_detections_to_trackers` function

  matched_indices = linear_assignment(-iou_matrix) 

  # standardize output to match sklearn implementation if using scipy
  if sk == False:
      row, col = matched_indices
      matched_indices = np.concatenate((row.reshape(-1,1), col.reshape(-1,1)), axis = 1)

Below code highlights the differences between the two implementations:

# sklearn implementation
from sklearn.utils.linear_assignment_ import linear_assignment
import numpy as np

cost = np.array([[4, 1, 3], [2, 0, 5], [3, 2, 2]])
result = linear_assignment(cost)
result # array([[0, 1],
       # [1, 0],
       # [2, 2]])

# scipy implementation
import numpy as np
from scipy.optimize import linear_sum_assignment as linear_assignment

cost = np.array([[4, 1, 3], [2, 0, 5], [3, 2, 2]])
result = linear_assignment(cost)
result # (array([0, 1, 2]), array([1, 0, 2], dtype=int64))

row, col = result
result = np.concatenate((row.reshape(-1,1), col.reshape(-1,1)), axis = 1)
result # array([[0, 1],
       # [1, 0],
       # [2, 2]], dtype=int64)

[rdzfv]

This happened to me when I changed:

• from sklearn.utils.linear_assignment_ import linear_assignment to
• from scipy.optimize import linear_sum_assignment as linear_assignment

Because sklearn's linear_assignment has been removed starting with version "0.23.0" and higher, I opted for Scipy's implementation of the function (which returns a tuple of np vectors as opposed to a matrix) (click here to learn more).

Thus, to get the equivalent result as if we were using sklearn's implementation I modified sort.py with below code:

import sklearn
skver = int(sklearn.__version__[2:4]) # extract middle number of version with format \d.\d\d.\d\d
if skver >= 23:
    from scipy.optimize import linear_sum_assignment as linear_assignment
    sk = False # we will not use sklearn implementation
else:
    from sklearn.utils.linear_assignment_ import linear_assignment
    sk = True # we will use sklearn implementation

# Below code was written after the `matched_indices` object had been created under the `associate_detections_to_trackers` function

  matched_indices = linear_assignment(-iou_matrix) 

  # standardize output to match sklearn implementation if using scipy
  if sk == False:
      row, col = matched_indices
      matched_indices = np.concatenate((row.reshape(-1,1), col.reshape(-1,1)), axis = 1)

Below code highlights the differences between the two implementations:

# sklearn implementation
from sklearn.utils.linear_assignment_ import linear_assignment
import numpy as np

cost = np.array([[4, 1, 3], [2, 0, 5], [3, 2, 2]])
result = linear_assignment(cost)
result # array([[0, 1],
       # [1, 0],
       # [2, 2]])

# scipy implementation
import numpy as np
from scipy.optimize import linear_sum_assignment as linear_assignment

cost = np.array([[4, 1, 3], [2, 0, 5], [3, 2, 2]])
result = linear_assignment(cost)
result # (array([0, 1, 2]), array([1, 0, 2], dtype=int64))

row, col = result
result = np.concatenate((row.reshape(-1,1), col.reshape(-1,1)), axis = 1)
result # array([[0, 1],
       # [1, 0],
       # [2, 2]], dtype=int64)

Sorry,I met the same problem. But I'm not quite understand where's sort.py and how to modify it?

[rdzfv]

This happened to me when I changed:

• from sklearn.utils.linear_assignment_ import linear_assignment to
• from scipy.optimize import linear_sum_assignment as linear_assignment

Because sklearn's linear_assignment has been removed starting with version "0.23.0" and higher, I opted for Scipy's implementation of the function (which returns a tuple of np vectors as opposed to a matrix) (click here to learn more).

Thus, to get the equivalent result as if we were using sklearn's implementation I modified sort.py with below code:

import sklearn
skver = int(sklearn.__version__[2:4]) # extract middle number of version with format \d.\d\d.\d\d
if skver >= 23:
    from scipy.optimize import linear_sum_assignment as linear_assignment
    sk = False # we will not use sklearn implementation
else:
    from sklearn.utils.linear_assignment_ import linear_assignment
    sk = True # we will use sklearn implementation

# Below code was written after the `matched_indices` object had been created under the `associate_detections_to_trackers` function

  matched_indices = linear_assignment(-iou_matrix) 

  # standardize output to match sklearn implementation if using scipy
  if sk == False:
      row, col = matched_indices
      matched_indices = np.concatenate((row.reshape(-1,1), col.reshape(-1,1)), axis = 1)

Below code highlights the differences between the two implementations:

# sklearn implementation
from sklearn.utils.linear_assignment_ import linear_assignment
import numpy as np

cost = np.array([[4, 1, 3], [2, 0, 5], [3, 2, 2]])
result = linear_assignment(cost)
result # array([[0, 1],
       # [1, 0],
       # [2, 2]])

# scipy implementation
import numpy as np
from scipy.optimize import linear_sum_assignment as linear_assignment

cost = np.array([[4, 1, 3], [2, 0, 5], [3, 2, 2]])
result = linear_assignment(cost)
result # (array([0, 1, 2]), array([1, 0, 2], dtype=int64))

row, col = result
result = np.concatenate((row.reshape(-1,1), col.reshape(-1,1)), axis = 1)
result # array([[0, 1],
       # [1, 0],
       # [2, 2]], dtype=int64)

I got to know that we were talking about the different project, so i opened a new issue https://github.com/yehengchen/Object-Detection-and-Tracking/issues/98#issue-860893050.

could you please give me a hand?

[SabraHashemi]

row, col = result

this worked for me,thank you

[Dipankar1997161]

If you encountered a problem in the linear_assignment.py try adding that extra second line.

Mine is for a different project, but the error was same for me.

indices = linear_assignment(cost_matrix) indices = np.hstack([indices[0].reshape(((indices[0].shape[0]), 1)),indices[1].reshape(((indices[0].shape[0]), 1))])

hope it helps :)