up vote 0 down vote favorite

I'm doing some image detection. The detection algorithm has a line of code that's causing an error.

The line of code:

im_rois = im_rois.astype(np.float, copy=False)

The error is:

Error: ValueError: setting an array element with a sequence.

I printed the contents of im_rois. They are:

[[ array([[  56.04,   57.32,  317.16,  285.16],
       [ 125.16,   17.64,  332.52,  285.16],
       [   1.  ,    1.  ,  427.24,  639.72],
       ..., 
       [ 201.96,    1.  ,  248.04,   50.92],
       [ 286.44,    1.  ,  350.44,   47.08],
       [ 304.36,  244.2 ,  338.92,  349.16]])]]

Curious if anyone can point me in the direction.

share|improve this question
1  
You don't seems to have a array but a list of list of array!! You can try im_rois = im_rois[0][0]. – Clodion Jul 26 '15 at 15:14
2  
I think it's best to ask the author of im_rois.astype. Either to provide more meaningful error, or debug their code. – luk32 Jul 26 '15 at 15:15
1  
Are you sure you haven't passed in an invalid argument to a function in the code you are using? The name im_rois suggests a sequence of "regions of interest", which (maybe) should be an array with shape (N, 4), where N is the number of RoIs. It looks like your im_rois is nested too deep. – Warren Weckesser Jul 26 '15 at 15:27
    
Thanks, yes its quite possible I've got some invalid inputs. The roi inputs are generated by another algorithm called selective search using matlab. The matlab file is loaded and converted into a roi array. I'm trying to figure out if I should go back and fix my matlab based code or persist with the matlab output I have. – gobob Jul 26 '15 at 15:33
1  
Is this matlab file a .mat? Do you load it with scipy loadmat? Matlab matrices load as numpy arrays, but matlab cells load as python lists (but may contains arrays). – hpaulj Jul 27 '15 at 16:53
up vote 2 down vote accepted

I can reproduce your error with this sequence:

In [1317]: im=[[np.array([[1,2,3]])]]  # list containing array

im.astype(float) would produce a AttributeError: 'list' object has no attribute 'astype' because it is a list, not an array.

But I can embed it in an object array:

In [1318]: x=np.empty((1,),dtype=object)
In [1319]: x[0]=im    
In [1320]: x
Out[1320]: array([[[array([[1, 2, 3]])]]], dtype=object)

The print looks like yours; note the commas in the inner list.

In [1321]: print(x)
[[[array([[1, 2, 3]])]]]

If it was a ndim array, the print wouldn't have the commas.

In [1325]: print(np.array([[[1,2,3]]]))
[[[1 2 3]]]

Now when I try a astype I get your value error.

In [1322]: x.astype(np.float)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-1322-1bc194707569> in <module>()
----> 1 x.astype(np.float)

ValueError: setting an array element with a sequence.

But without knowing how im_rois is created I can't suggest ways of avoiding this. If im_rois were

array([[  56.04,   57.32,  317.16,  285.16],
       [ 125.16,   17.64,  332.52,  285.16],
       [   1.  ,    1.  ,  427.24,  639.72],
       ...]])

it wouldn't give this error.

I'd check to make you sure you are feeding this package the right kinds of input, whether it be a list, an array or a scalar.

vstack is able to remove this intermediate 'object' layer, allowing astype.

In [1343]: np.vstack(x)
Out[1343]: array([[[[1, 2, 3]]]])
share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.