Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• The proposed framework, 'Only What's Necessary', combines breadth-based and depth-based data minimization mechanisms to suppress PII while preserving cues relevant to anomaly detection.method 95%
• Temporal sampling with a stride of 5 (s=5) results in an AUC of 72.98% for anomaly detection, while temporal sampling with a stride of 10 (s=10) results in an AUC of 67.49%.result 90%
• Masking as a depth-based minimization technique results in an AUC of 57.25% for anomaly detection.result 90%
• Combining temporal sampling (s=5) with masking results in an AUC of 55.81% for anomaly detection.result 90%
• The paper employs Pareto analysis to characterize the trade-off between privacy and utility in video anomaly detection.method 95%
• The CUHK Avenue dataset contains 16 training and 21 testing videos, captured using stationary cameras with fixed viewpoints.technical 90%
• The PAHMDB dataset is a subset of HMDB51 and comprises 515 videos used for both training and evaluation, providing frame-level annotations for privacy-sensitive attributes.technical 90%
• The proposed framework, 'Only What's Necessary', combines breadth-based and depth-based data minimization mechanisms to suppress PII while preserving cues relevant to anomaly detection.method 95%
• Temporal sampling with a stride of 5 results in an AUC of 72.98% for anomaly detection, while temporal sampling with a stride of 10 results in an AUC of 67.49%.result 90%
• Masking as a depth-based minimization technique results in an AUC of 57.25% and a cMAP of 59.10%.result 90%
• Combining temporal sampling (s=5) with masking results in an AUC of 55.81% and a cMAP of 57.23%.result 90%
• The framework employs Pareto analysis to characterize the trade-off between privacy and utility, identifying 'sweet spot' operating points.method 95%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card