IPAB Workshop - 30/7/26 Speaker: Rim Rubavicius Title: Robots are not the Enemy Abstract: You can get a PhD in Robotics without ever touching a physical robot. This often arises due to various time, logistic and feasibility constraints. In this talk, I will argue that working with "real robots" should be seen as a research opportunity and not as a hurdle. I will do this, by reflecting on three of my personal case studies of using and deploying real robots for teaching and public engagement activities and how this experience has shaped my thinking on what matter in developing the science of robotics. Speaker: Rohit Bharadwaj Title: StablePrivacy: Zero-Shot Full-Body Video Anonymization Using Diffusion Models For Downstream Data Utility Abstract: Existing privacy research mainly targets facial anonymization in images, overlooking identifying cues beyond the face (\eg, clothing and hair) or the challenges of video data. We propose \textbf{StablePrivacy}, a diffusion-based framework for full-body video anonymization that improves the privacy--utility trade-off. Built on Stable Diffusion, our method operates in a zero-shot manner without dataset-specific training. The framework preserves pose and scene context via ControlNet conditioning and latent-space inpainting, while a memory bank enforces temporal consistency across frames. We further introduce an injection module that retains task-relevant structure while removing identity cues. Experiments on video instance segmentation, action recognition, and person re-identification demonstrate that StablePrivacy better preserves downstream task performance than existing full-body anonymization methods while ensuring privacy. The code will be released publicly. Jul 30 2026 13.00 - 14.00 IPAB Workshop - 30/7/26 Rim Rubavicius & Rohit Bharadwaj G.07 This article was published on Monday 27 July 2026
IPAB Workshop - 30/7/26 Speaker: Rim Rubavicius Title: Robots are not the Enemy Abstract: You can get a PhD in Robotics without ever touching a physical robot. This often arises due to various time, logistic and feasibility constraints. In this talk, I will argue that working with "real robots" should be seen as a research opportunity and not as a hurdle. I will do this, by reflecting on three of my personal case studies of using and deploying real robots for teaching and public engagement activities and how this experience has shaped my thinking on what matter in developing the science of robotics. Speaker: Rohit Bharadwaj Title: StablePrivacy: Zero-Shot Full-Body Video Anonymization Using Diffusion Models For Downstream Data Utility Abstract: Existing privacy research mainly targets facial anonymization in images, overlooking identifying cues beyond the face (\eg, clothing and hair) or the challenges of video data. We propose \textbf{StablePrivacy}, a diffusion-based framework for full-body video anonymization that improves the privacy--utility trade-off. Built on Stable Diffusion, our method operates in a zero-shot manner without dataset-specific training. The framework preserves pose and scene context via ControlNet conditioning and latent-space inpainting, while a memory bank enforces temporal consistency across frames. We further introduce an injection module that retains task-relevant structure while removing identity cues. Experiments on video instance segmentation, action recognition, and person re-identification demonstrate that StablePrivacy better preserves downstream task performance than existing full-body anonymization methods while ensuring privacy. The code will be released publicly. Jul 30 2026 13.00 - 14.00 IPAB Workshop - 30/7/26 Rim Rubavicius & Rohit Bharadwaj G.07 This article was published on Monday 27 July 2026
