Summary of Badhmp: Backdoor Attack Against Human Motion Prediction, by Chaohui Xu et al.
BadHMP: Backdoor Attack against Human Motion Prediction
by Chaohui Xu, Si Wang, Chip-Hong Chang
First submitted to arxiv on: 29 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed BadHMP attack targets human motion prediction models by generating poisoned training samples that embed a localized backdoor trigger in the skeleton. This trigger causes specific joints to remain still or follow predefined motions, allowing the attacker to control the predicted future sequences. The approach involves selecting suitable samples for poisoning and designing triggers that guarantee smoothness and naturalness while evading detection. Experimental results on two datasets (Human3.6M and CMU-Mocap) and two network architectures (LTD and HRI) demonstrate the effectiveness, stealthiness, and high-fidelity of BadHMP. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary BadHMP is a new kind of attack that can trick machines that predict human movements from past observations. This is important because some systems rely on these predictions to make decisions. The bad guys can create fake training data that makes the machine think certain joints should be still or moving in specific ways, allowing them to control what the machine predicts. The creators of BadHMP tested it on two different datasets and showed that it works well. |
