Real-Time Privacy Preservation for Robot Visual Perception
Authors: Minkyu Choi, Yunhao Yang, Neel P. Bhatt, Kushagra Gupta, Sahil Shah, Aditya Rai, David Fridovich-Keil, ufuk topcu, Sandeep P. Chinchali
TMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Quantitative evaluations show that PCVS achieves over 95 percent specification satisfaction rate in multiple datasets, significantly outperforming other methods. The satisfaction rate is consistently above the theoretical bounds across all datasets, indicating that the established bounds hold. Additionally, we deploy PCVS on robots in real-time operation and show that the robots operate normally without being compromised when PCVS conceals objects. |
| Researcher Affiliation | Academia | Minkyu Choi1 , Yunhao Yang1 , Neel P. Bhatt1 , Kushagra Gupta 1, Sahil Shah1, Aditya Rai1, David Fridovich-Keil1, Ufuk Topcu1, Sandeep P. Chinchali1 1 The University of Texas at Austin |
| Pseudocode | Yes | Algorithm 1: Real-Time Video Abstraction |
| Open Source Code | No | The paper does not contain any explicit statements about releasing source code, nor does it provide links to a code repository. |
| Open Datasets | Yes | We use YOLOv9 (Wang et al., 2025) on large-scale image datasets Image Net (Deng et al., 2009) and MS COCO (Lin et al., 2014) and a real-world driving dataset UFPR-ALPR (Laroca et al., 2018). |
| Dataset Splits | No | The paper describes how evaluation datasets (ED1, ED2, ED3) were constructed (e.g., "five different video lengths: 10, 25, 50, 100, and 200, with 25 video samples for each duration" for ED1), but it does not provide specific train/test/validation splits for model training or how the underlying models used were split. |
| Hardware Specification | Yes | The experiments are performed using the video collected by the ground robot with an Apple M2 CPU. ... Furthermore, we tested our method with YOLOv9 on both Intel Xeon Gold CPU and Nvidia A5000 GPU. |
| Software Dependencies | No | The paper mentions software like YOLOv9, Yolov9e, Yolov8x-worldv2, Faster RCNN, GPT-4 Vision, and Video LLa VA, but it does not provide specific version numbers for these, only citations to their respective papers. |
| Experiment Setup | Yes | The probabilistic guarantee of privacy preservation throughout the complete operation is 0.81. ... PCVS effectively maintains privacy above the user-given privacy threshold of 0.80, denoted as PGk(Φ) > 0.80. ... Table 1: The number of iterations, latency, and PSNR remain nearly identical for different thresholds. |