search

maestrojeong / deformable_convnet

Tensorflow implementation of deformable convnet on MNIST
29 stars 16 forks source link

question about the input with unknown size. #2

Open wensihan opened 5 years ago

wensihan commented 5 years ago

Hello, if I set the input image as [1, None, None, 3] instead of [-1,28,28,1], there will come out to the following bugs:

offset_map = tf.reshape(offset_map, [batch, i_h, i_w, f_h, f_w, 2]) TypeError: Failed to convert object of type <class 'list'> to Tensor. Contents: [1, None, None, 5, 5, 2]. Consider casting elements to a supported type.

Do you have any advice? Looking to your reply.

maestrojeong commented 5 years ago

My code doesn't support [1, None, None, 3] shape, since I get the shape of tensor like 'a, b, c, d = tensor.shape' if shape of 'tensor' contains 'None', my code makes an error. Try with specified shape.

RituYadav92 commented 4 years ago

@wensihan Hey, were you able to apply deformable conv in keras successfully? I am trying to this in an object detection task from last few days and facing error due to None shape(which is quite common). Please let me know if you found any solution to this issue. Thanks and regards, Ritu Yadav