To Keystone or Not to Keystone, that is the Correction
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
To Keystone or not to Keystone, that is the correction... and indeed the question! Outside of highly constrained conditions, the vast majority of photographed imagery of the natural environment is taken non-square to the objects that they represent Consequently, those objects appearing at a distorted perspective may be computationally corrected via Keystone Correction. This disparity is frequently observed when considering imagery sourced from vehicle-mounted cameras, such as those levied in autonomous vehicle infrastructure or by streetscape collection initiatives such as Google Street View. As visual creatures, the lived environment proximal to roadways is filled with text- and numeric-based advertisements vying for our attention and, conveniently, this signage isn't placed perpendicular to a vehicle's forward-facing camera. Given the perspective distortion of the text and/or values contained therein, their automated detection and reading may benefit from Keystone correction. In this work, we address the yet-unanswered question: what benefit might we expect from Keystone correction preprocessing of images? We do not explicitly promote the use of Keystone correction but rather, evaluate its utility within a prediction pipeline. To this end, we leverage the Gas Prices of America (GPA) dataset containing multi-digit, multi-price values and the French Street Sign Names (FSNS) multi-word text dataset given their known geometry enabling the automation of image Keystone correction. We compare the outcomes of Keystoned imagery versus non - Keystoned imagery along five axes: 1) predictive performance, 2) annotation correctness, 3) algorithmic computational complexity and empirical time estimation, 4) image scaling, and 5) degree of perspective transform. From our findings, we arrive at several recommendations on both the benefit & burden of Keystone correction to inform future research on extracting information in the wild.
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.
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 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.001 |
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 it