search

ethz-asl / hand_eye_calibration

Python tools to perform time-synchronization and hand-eye calibration.
BSD 3-Clause "New" or "Revised" License
430 stars 115 forks source link

Experiments package and updated scripts #48

Closed mfehr closed 7 years ago

mfehr commented 7 years ago

Depends on: https://github.com/ethz-asl/hand_eye_calibration/pull/49

This PR updates moves all experiment scripts into a separate package and updates the main script such that it now runs the complete pipeline (with optimization) based on the following configuration matrix:

Method without filtering and without optimization without filtering and with optimization with filtering and without optimization with filtering and with optimization
Baseline
RANSAC classic
RANSAC scalar inliers
Exhaustive search classic - -
Exhaustive search scalar inliers - -

FYI: @HannesSommer

HannesSommer commented 7 years ago

I've merged #49 . Strange that github doesn't seem to update the diff w.r.t. master. My changes are still in there. That's a bit bad for the review I suppose. If that remains I suggest to recreate the PR.

HannesSommer commented 7 years ago

About the todo "add mode: optimization only with initial guess for time alignment":

But what should the initial guess be for the the relative pose? I must initialize somehow. Were you actually thinking about identity (in SE3)? That would be easy but also quite unfair for different experiments I reckon. One could use a spoiled initial guess. Something like always off a certain amount of the best we know.

ffurrer commented 7 years ago

I've merged #49 . Strange that github doesn't seem to update the diff w.r.t. master. My changes are still in there. That's a bit bad for the review I suppose. If that remains I suggest to recreate the PR.

You have to merge in master again (after the merge), such that this disappears, did so now.

HannesSommer commented 7 years ago

Really? Interesting. Does somebody understand that? I don't see any other possible causal connection other than triggering a update of the diff by causing an update of the ref. I'd love to quickly try the inverse (so undo @ffurrer action with a force push and see what happens). Is somebody against that?

ffurrer commented 7 years ago

@HannesSommer feel free to force push ;)

HannesSommer commented 7 years ago

Thanks. Yep, identical history + better diff :) => Lazy github!

mfehr commented 7 years ago

@HannesSommer @ffurrer I ran the experiment over night and it succeeded except for the optimization-only mode, there is still a bug in the script. If you haven't started a review, maybe you should wait a bit, if you have it's ok as well, the changes shouldn't be too extensive.

ffurrer commented 7 years ago

I started, but will hence wait with the rest :) And nice that it seems to be almost working :)

HannesSommer commented 7 years ago

Always it is my stuff that fails :(. Please tell me if I can help with the fixing. How does it fail?

HannesSommer commented 7 years ago

@mfehr , please wait for PR https://github.com/ethz-asl/oomact/pull/11 with the verification whether relative measurements were actually activated. After that the INFO log should actually contain an output about the effect. You should be able to find it by searching for absoluteMeasurements.

HannesSommer commented 7 years ago

I've just merged it.

mfehr commented 7 years ago

@ffurrer should be ready now, running the complete experiment again now.

mfehr commented 7 years ago

@ffurrer thx a lot for reviewing this huge PR!