Bug where np.round returns numpy.float64

[PilleOtp]

Hi,

This pull request fixes a bug that caused a TypeError in the setKpParams function of the cocoeval.py file when using Python 3.7 and NumPy 1.21.6. The error occurred due to the use of np.linspace with a numpy.float64 object as the number of sample points.

Error details:

File "path/to/anaconda3/envs/HRNet2/lib/python3.7/site-packages/pycocotools-2.0-py3.7-macosx-10.9-x86_64.egg/pycocotools/cocoeval.py", line 556, in setKpParams
    self.iouThrs = np.linspace(.5, 0.95, np.round((0.95 - .5) / .05) + 1, endpoint=True)
  File "<__array_function__ internals>", line 6, in linspace
  File "path/to/anaconda3/envs/HRNet2/lib/python3.7/site-packages/numpy/core/function_base.py", line 120, in linspace
    num = operator.index(num)
TypeError: 'numpy.float64' object cannot be interpreted as an integer

Environment:

Python: 3.7
NumPy: 1.21.6

To resolve this issue, I have modified the np.linspace call in the setKpParams/setDetParams function to ensure that an integer is provided as the number of sample points. This change should maintain compatibility with other NumPy versions as well.