Privacy Through the Looking Glass: Diminished Reality to Enhance Privacy Beyond Reflective Leaks
Bibliographic record
Abstract
In the wake of recent major privacy breaches and stringent regulations such as the GDPR, protecting privacy has become more critical than ever. Privacy concerns are exacerbated by the spread of pervasive technologies like Mixed Reality (MR), which have cameras continuously collecting visual data from spaces around us. While direct sensitive content that appears in a feed seems like the only problem, sensitive information might leak indirectly from other subtle visual cues, particularly reflections. Although several privacy-preserving frameworks have emerged, they are often two-dimensional, sensitive to motion, and overall not tailored for MR environments where private objects could be inadvertently revealed within a dynamic three-dimensional space. Furthermore, none of the frameworks addresses the issue of reflective privacy leakage. We propose a three-dimensional visual privacy-preserving Diminished Reality (DR) framework that obfuscates private objects and their reflections during non-trivial motion. The private object obfuscation is governed by a machine learning model that predicts the potential failure of the algorithm in real time, ensuring complete obfuscation when necessary to prevent privacy leaks. As for reflection obfuscation, we introduce a reflection detection and obfuscation stage into the system to further prevent privacy leaks, even from indirect sources such as mirrors or reflective surfaces. Our evaluation demonstrates that we effectively increase both privacy and privacy-utility tradeoff levels compared to state-of-the-art, reducing privacy failures by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5 4. 2 \%}$</tex>, and improving the tradeoff by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 1 \%}$</tex>.
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 | 0.002 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.005 |
| Science and technology studies | 0.002 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.007 | 0.006 |
| Research integrity | 0.000 | 0.001 |
| 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".