philjd / tf-quaternion Goto Github PK

Hi,
I am trying to normalize the quaternion, but I encountered this error.

q = tfq.Quaternion([0.1, 0.5, 0.1, 0.001]).normalized()

TypeError Traceback (most recent call last)
in
----> 1 q = tfq.Quaternion([0.1, 0.5, 0.1, 0.001]).normalized()
2 q

/disk/anaconda3/lib/python3.7/site-packages/tfquaternion/tfquaternion.py in scoped_func(*args, **kwargs)

/disk/anaconda3/lib/python3.7/site-packages/tfquaternion/tfquaternion.py in normalized(self)

/disk/anaconda3/lib/python3.7/site-packages/tfquaternion/tfquaternion.py in scoped_func(*args, **kwargs)

/disk/anaconda3/lib/python3.7/site-packages/tfquaternion/tfquaternion.py in abs(self, keepdims)

/disk/anaconda3/lib/python3.7/site-packages/tfquaternion/tfquaternion.py in scoped_func(*args, **kwargs)

/disk/anaconda3/lib/python3.7/site-packages/tfquaternion/tfquaternion.py in norm(self, keepdims)

/disk/anaconda3/lib/python3.7/site-packages/tensorflow/python/util/traceback_utils.py in error_handler(*args, **kwargs)
151 except Exception as e:
152 filtered_tb = _process_traceback_frames(e.traceback)
--> 153 raise e.with_traceback(filtered_tb) from None
154 finally:
155 del filtered_tb

/disk/anaconda3/lib/python3.7/site-packages/tensorflow/python/util/dispatch.py in op_dispatch_handler(*args, **kwargs)
1074 if iterable_params is not None:
1075 args, kwargs = replace_iterable_params(args, kwargs, iterable_params)
-> 1076 result = api_dispatcher.Dispatch(args, kwargs)
1077 if result is not NotImplemented:
1078 return result

TypeError: Got an unexpected keyword argument 'keep_dims'

When using a tensor of shape [N, 4], the Quaternion.abs function produces a scalar. I would expect this to behave more like the Quaternion.norm function, which does preserve dimensions. An unfortunate side effect of this is that the Quaternion.normalized scales each quaternion by the same scalar, instead of by the norm of each quaternion. Is my understanding of the expected functionality correct?

i try to rotate vectors with an array or quaterion, seems not working for me

EDIT: Given the definition of normalized quaternion, shouldn't norm() return self._q / self.abs(), because I was expecting the normalized quaternion when I called norm() and I am sure other people might make that mistake, given that you have a different method "abs()" for the actual norm of a quaternion

I am getting the following Error AttributeError: 'numpy.ndarray' object has no attribute 'normalized'. could somone help? thank you in advance

Hi PhilJD,

May I know what Tensorflow version does this library supports? Thanks

Thanks for the module but it is not compatible with TF1.13 (stable) or TF2.0 (alpha) due to the keyword argument keep_dims being deprecated. Please use keepdims in functions like reduce_sum().

Hello and thank you for this awesome repo.

I am quite certain that this is not a direct issue with this repo, but more or less a tensorflow issue which arises here. But I hope that somebody might help me with it...

If one does the following:

q1 = tfq.Quaternion(y[:-3])

and y is a 6 element vector then the error will come:
ValueError: Can't create a quaternion form a tensor with shape (?,6). The last dimension must be 4.

So I am wondering what would be the best way to slice an array to 4 element vector which can then be passed to the lib?

I tried it with creatin a variable, however this results in a whole other complication, since the shape must be known beforehand,... (see here: https://stackoverflow.com/questions/51493490/validate-shape-equivalent-in-keras-in-comparison-to-tensorflow )

I am trying to write a metrics using quaternions, the code looks like the follwoing:

def vectorMAEquat(y_true,y_pred):
        
    p1 = tf.constant([[1,0,0],[0,1,0],[0,0,1]],dtype=tf.float32)
    p2 = tf.constant([[1,0,0],[0,1,0],[0,0,1]],dtype=tf.float32)
  
    q1 = tfq.Quaternion(y_true)
    q2 = tfq.Quaternion(y_pred)
    
    rot1 = tf.matmul(q1.as_rotation_matrix(),p1)
    rot2 = tf.matmul(q2.as_rotation_matrix(),p2)
    
    return K.mean(K.abs(rot1 - rot2))

However, this gives me the error: ValueError: Cannot infer num from shape(?,?) which comes form the function

w,x,y,z = tf.unstack(self.normalized().value(),axis=-1) from the file tfquaternion.py , in line 364 in the function as_rotation_matrix.

A little bit of googling brought me here: https://stackoverflow.com/questions/45404056/tf-unstack-with-dynamic-shape

So, maybe this would be a nice feature, if as_rotation_matrix could work with dynamic_partition, but I am not fluent enough in tensorflow to achieve this, so any pointers would be really appreciated.

Thank you