Hello,

First, thank you for sharing your source code of your SS-DPC-Net research!

I'm planning to use it on our own dataset instead of the KITTI dataset and I'm wondering what the required files are in order to successfully run your source code. I'll greatly appreciate any tips you can give me.

Best regards,

Paul

Thank you for sharing your nice work!
I have a question about SE(3) pose concatenation in your code implementations.

Let's define the following symbol:
$T^w_t$: SE(3) pose at time $t$ w.r.t. $w$.

Then, as I know, the relative dT pose from $t-1$ to $t$ and pose concatenation should be as follows:

$T^{t-1}_{t} $

$=( T^w_{t-1} )^{-1} T^{w}_{t}$

and

$T^{w}_{t}$

$= T^w_{t-1} T^{t-1}_t$

However, according to the your code implementations of get relative pose and pose concat. using dT pose,

they are defined as following reverse order:

$T^{t-1}_t$

$= (T^w_{t})^{-1} T^w_{t-1}$

and

$T^{w}_{t}$

$= (T^{t-1}_{t} {(past pose)}^{-1})^{-1}$

where
$(past pose)$

$= (T^{w}_{t-1})$

these are somewhat different from my definition, and my pdf visualization result of VO estimation is weird.
Why is it implemented in reverse order?

Thank you

The error calculation is very confusing for me, more specifically the test trajectory function in validate.py. I am trying to run the network with non-kitti dataset, and I have made everything to look like kititi dataset, however I am getting a wrong error value between the GT and pose estimate. The value is around 7m in translation, and I used evo and actually the result is around 0.1m, any idea what might be the reason?