search

alexgkendall / caffe-posenet

Implementation of PoseNet
Other
496 stars 204 forks source link

Where does normalization of the output quaternion happen? #6

Closed alykhantejani closed 8 years ago

alykhantejani commented 8 years ago

Hi @alexgkendall

I've been looking through the train_posenet.prototxt and I can't see where the normalization of the predicted quaternion (from cls3_fc_pose_wpqr) happens. I see that the output of cls3_fc_pose_wpqr and the label go into the Euclidean loss (with weight beta), but in the paper equation 2 shows that the output quaternion is made a unit vector.

Am I missing something here?

Thanks, Aly

alexgkendall commented 8 years ago

Hey Aly,

That's correct - this functionality is not implemented in this Caffe code. This is something which I have in my Torch implementation of PoseNet. In practice, I have found it not to make a significant difference in the network's performance so I have omitted it here. Sorry for the confusion.

You can see in the evaluation scripts that I normalise the resulting quaternion vector for testing.

alykhantejani commented 8 years ago

Hi @alexgkendall,

Thanks for the clarification. Is your torch implementation public yet (with/without pre-trained models)?

Cheers, Aly

alexgkendall commented 8 years ago

No, not at this stage.

eeshandhekane commented 7 years ago

Hi @alexgkendall !

Is the Torch implementation of PoseNet public yet?