Using tracks in translation averaging #594
Conversation
|
Translation averaging metrics: Data association metrics: Bundle adjustment metrics: |
|
All places where we see a negative impact are already small numbers in terms of absolute value. |
|
Latency for translation averaging is expected to increase by 2-3x. This is dominated by an increase in the outlier rejection time as we have more measurements. |
| def __get_initial_values(self, wTi_initial: List[Optional[PosePrior]]): | ||
| # Compute outlier weights using MFAS. | ||
| # TODO(ayush): parallelize this step. | ||
| outlier_weights: List[Dict[Tuple[int, int], float]] = [] |
There was a problem hiding this comment.
Just curious: can we make outlier_weights dict keyed on symbols instead of integers?
There was a problem hiding this comment.
it is already. I dont know why there is an int type hint. are Symbols ints in python?
| for (i1, i2), i2Ui1 in i2Ui1_dict.items(): | ||
| wRi2 = wRi_list[i2] | ||
| if i2Ui1 is not None and wRi2 is not None: | ||
| # TODO: what if wRi2 is None, but wRi1 is not? Can we still transform. |
There was a problem hiding this comment.
Why are we enforcing both cameras to have a global rotation. It makes sense to do it, but why do it now?
|
Merging this, I think all comments have been addressed, thanks for the review! |
We have the 2D tracks before translation averaging. These essentially give us 3D direction measurements from cameras to 3D points. Following 1DSfM, we add these to the translation averaging problem to provide more constraints.
We only add a few tracks (this number needs to be tuned).
This PR also refactors translation averaging, moves some functions to utils, and adds unit tests.