Can I make a layer of which input is disparity & output is Rt?

ankurhanda / gvnn

gvnn: Geometric Vision with Neural Networks

445 stars 71 forks source link

Can I make a layer of which input is disparity & output is Rt? #20

Open SeokjuLee opened 6 years ago

SeokjuLee commented 6 years ago

With a given disparity map between i1 and i2, I want to generate the relative pose. Can I make this layer with GVNN library? If it is possible, could you give me an advice? Thank you!

ankurhanda commented 6 years ago

you can take inspiration from this https://github.com/ankurhanda/gvnn/blob/master/imagewarpingSE3.lua

SeokjuLee commented 6 years ago

@ankurhanda Ok Thank you so much!

SeokjuLee commented 6 years ago

@ankurhanda Thanks for your inspiration, but still I'm not sure how to implement the function of "input: disparity, output: R|T". I think imagewarpingSE3 recieves two images and outputs SE3 which is transpose matrix. Is this code just an operation or a trainable network? Could you give me some advices?