Vfrdtyky

When I convert two lists of numpy arrays to numpy arrays of numpy arrays, something confused happened.
The first list X_s changed to a numpy array with shape of (1980, 384, 448, 1), which is good for training, but the second list X_l chaned to a numpy arrays with shape of (2013,).
I check their dtype, and the first become float64 while the second become object of numpy array.
Why this happened?

print(len(X_s)) # 1980

print(len(X_l)) # 2013

print(X_s[0].dtype)  # float64

print(X_l[0].dtype)  # float64

print(X_s[0].shape)  # (384, 448, 1)

print(X_l[0].shape)  # (384, 448, 1)



for i in range(len(X_l)): 

    X_l[i] = np.array(X_l[i], dtype = np.float64)

for i in range(len(X_s)):

    X_s[i] = np.array(X_s[i], dtype = np.float64)



X_s = np.array(X_s)

X_l = np.array(X_l)



print(type(X_s[0]))  # <class 'numpy.ndarray'>

print(type(X_l[0]))  # <class 'numpy.ndarray'>



print(X_s.dtype) #  flaot64

print(X_l.dtype) #  object

print(X_s.shape) # （1980, 384, 448, 1）

print(X_l.shape) # （2013，）

After added two for loops to make sure the elements are in uniform type, nothing changed.

It looks very likely that the elements of the original X_l list are not of uniform type. (You only show us the type of the first element but not the rest.)

When NumPy tries to convert that list to an array, it notices that and coerces everything to object.

Demo:

In [10]: X_s = [np.array([1]), np.array([2])]



In [11]: X_l = [np.array([1]), 2]



In [12]: np.array(X_s)

Out[12]:

array([[1],

       [2]])



In [13]: np.array(X_l)

Out[13]: array([array([1]), 2], dtype=object)

(This example is made up but consistent with your observations.)