Bibliographic record
Abstract
Place recognition is the online task of detecting revisits to previously seen locations and is a key to many navigational systems. In Simultaneous Localization and Mapping, recovering the relative camera pose between recognized visit and revisit (e.g. using bundle adjustment) allows for global map optimization, improving localization accuracy. Visual SLAM recovers structure to estimate camera movement but it is typically not used for visual place recognition. Limited past work which adapted LiDAR place recognition descriptors to SLAM-recovered physical structure found superior robustness to visual effects vs appearance-based VPR, but overall had poorer recall. It was found that LiDAR descriptors’ whole-scan matching assumes excellent 360 degree pointcloud coverage while cameras have limited FoV. We observe that SLAM-tracked points congregate on objects and distinct elements, resulting in sparsity that impacts whole-scan matching. To us this also suggests use of clustering to extract these aggregate congregations as landmarks whose configuration can be matched. Exploring this approach we found that the landmarks generated still vary in detected position, but a far more significant hurdle is that the same landmarks may not be repeatedly clustered each time a scene is visited. This is due to large-scale clustering still being sensitive to instability in the individual SLAM points. This was improved significantly but not sufficiently through visual semantic labeling of the initial 3D points, helping to provide more stable, guided clustering solutions. Still, single missing or “outlier” landmarks are detrimental to successful association between landmark sets. To address this instability in future work we recommend careful selection of salient points from those collected by SLAM, for those which can be expected to be the most stable and repeatably detected. This is expected to provide more stable landmarks than large-scale clustering of detected points which relies on a center-of-mass approach.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".