Make bundle adjustment deterministic and robust to outliers#5
Draft
jasper-tms wants to merge 1 commit into
Draft
Make bundle adjustment deterministic and robust to outliers#5jasper-tms wants to merge 1 commit into
jasper-tms wants to merge 1 commit into
Conversation
Frame subsampling is now seeded (np.random.default_rng) and uses np.random.choice(replace=False) instead of np.random.randint, which had duplicates and silently excluded the last frame. least_squares now uses loss='soft_l1' with f_scale=8 px so bad 2D detections don't dominate the fit. max_nfev raised from 100 to 500 since the gentler robust gradient needs more iterations to converge.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Bundle adjustment used to be stochastic - each time it runs, it would pull 500 random images out of the dataset and do bundle adjustment on those. If some of the images in the dataset had poor 2D keypoint estimation, the results of bundle adjustment could be wildly different depending on whether the outlier keypoint frames were in or not in the selected 500 images. We make two fixes:
I need to test this a bit more before confirming that it constitutes a real improvement, but I think it is.