mingyuyng/deepvo

Unofficial implementation of "Deepvo: Towards end-to-end visual odometry with deep recurrent convolutional neural networks"

deepvo's Issues

Request a predicted pose file

Thank you for the code you submitted. Can you provide the predicted pose file for reference? Thank you!

Changing rnn to GRU causes test error.

Hi @mingyuyng. Thankyou for the work.

I made a couple changes to the parameters and everything worked fine for both training and testing.

I changed the RNN to GRU. The training was fine, but for testing, it leads to this error:

Traceback (most recent call last):
  File "testGRU.py", line 114, in <module>
    test(par.test_video)
  File "testGRU.py", line 62, in test
    angle, trans, hc = M_deepvo.forward(x_in, prev=prev)
  File "xxx/modelGRU.py", line 109, in forward
    out, hc = self.rnn(x) if not prev else self.rnn(x, prev)
_RuntimeError: bool value of Tensor with more than one value is ambiguous_

I tried looking in other places, but in those cases it was because of how the loss was calculated, but in this case the error occurs when the forward method is called.

I am new to ML and don't even know where to get started with this. Any help would be appreciated.