Composing Driving Worlds through Disentangled Control for Adversarial Scenario Generation

Dimensions overall score 8.0

• CompoSIA is a compositional driving video simulator that disentangles scene structure, object identity, and ego actions.method 70%
  Evidencepartial

  We introduce CompoSIA, a compositional driving video simulator that disentangles these traffic factors, enabling fine-grained control over diverse adversarial driving scenarios.

  Implicationpartial

  This is a core statement of the paper's contribution, explicitly mentioned in the abstract.

  Verificationpartial

  partial

• A noise-level identity injection method allows for pose-agnostic identity generation from a single reference image.method 70%
  Evidencepartial

  To support controllable identity replacement of scene elements, we propose a noise-level identity injection, allowing pose-agnostic identity generation across diverse element poses, all from a single reference image.

  Implicationpartial

  This describes a specific technical innovation introduced by the paper, clearly stated in the abstract.

  Verificationpartial

  partial

• CompoSIA achieves a 17% improvement in FVD for identity editing compared to state-of-the-art baselines.result 70%
  Evidencepartial

  Extensive comparisons demonstrate superior controllable generation quality over state-of-the-art baselines, with a 17% improvement in FVD for identity editing and reductions of 30% and 47% in rotation and translation errors for action control.

  Implicationpartial

  This is a specific quantitative result demonstrating the superiority of the proposed method.

  Verificationpartial

  partial

• The proposed hierarchical dual-branch action control mechanism reduces rotation and translation errors by 30% and 47% respectively.result 70%
  Evidencepartial

  Extensive comparisons demonstrate superior controllable generation quality over state-of-the-art baselines, with a 17% improvement in FVD for identity editing and reductions of 30% and 47% in rotation and translation errors for action control.

  Implicationpartial

  This is a specific quantitative result detailing the performance improvement of the action control mechanism.

  Verificationpartial

  partial

• Adversarial scenario synthesis using CompoSIA leads to a 173% average increase in collision rate for downstream planners.result 70%
  Evidencepartial

  Furthermore, downstream stress-testing reveals substantial planner failures: across editing modalities, the average collision rate of 3s increases by 173%.

  Implicationpartial

  This is a significant result demonstrating the effectiveness of the generated adversarial scenarios in stressing autonomous driving planners.

  Verificationpartial

  partial

• Current controllable generative models provide incomplete or entangled guidance for autonomous driving scenario synthesis.limitation 70%
  Evidencepartial

  Synthesizing these scenarios is crucial, yet current controllable generative models provide incomplete or entangled guidance, preventing the independent manipulation of scene structure, object identity, and ego actions.

  Implicationpartial

  This statement sets up the problem that the paper aims to solve and is directly mentioned in the abstract.

  Verificationpartial

  partial

• The proposed noise-level identity injection enables pose-agnostic identity generation.technical 70%
  Evidencepartial

  allowing pose-agnostic identity generation across diverse element poses, all from a single reference image.

  Implicationpartial

  This is a specific technical capability of the proposed method, directly stated.

  Verificationpartial

  partial

Competitive landscape

CompoSIA is a compositional driving video simulator that enables fine-grained control over adversarial driving scenarios to enhance safety in autonomous driving.