Research map / 3D vision / 3D generation
Hermes++: Toward a Unified Driving World Model for 3D Scene Understanding and Generation
Xin Zhou, Dingkang Liang, Xiwu Chen, Feiyang Tan, Dingyuan Zhang, Hengshuang Zhao, Xiang Bai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add L_cos implicit geometric regularization (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | Chamfer Distance 3s: 1.637 -> 1.441 | Improved |
| add both L_cos and L_gram (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | Chamfer Distance 3s: 1.637 -> 1.436 | Improved |
| add L_gram alone (no L_cos) (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | Chamfer Distance 3s: 1.637 -> 1.544 | Improved |
| add Current-to-Future Link (simple link) (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | Chamfer Distance 3s: 2.377 -> 1.542 | Improved |
| add Textual Injection to the Link (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | Chamfer Distance 3s: 1.542 -> 1.506 | Improved |
| add Ego Modulation to the Link (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | Chamfer Distance 3s: 1.506 -> 1.442 | Improved |
| add deep task interaction (unified vs separated unification) (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | CIDEr: 0.703 -> 0.720 | Improved |
| add world queries bypassing the LLM (setting b vs a) (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | Chamfer Distance 3s: 1.634 -> 1.526 | Improved |
| initialize world queries from BEV features by max pooling instead of random init (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | Chamfer Distance 3s: 1.448 -> 1.436 | Improved |
| use downsample x4 instead of x8 for BEV (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | Chamfer Distance 3s: 1.781 -> 1.436 | Improved |
| use downsample x4 instead of direct query x4 (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | Chamfer Distance 3s: 2.012 -> 1.436 | Improved |
| use 4 world queries instead of none (ablation) | Hermes++ (1.8B) · OmniDrive-nuScenes val | Chamfer Distance 3s: 1.478 -> 1.436 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ViDAR — ViDAR: Video Diffusion-Aware 4D Reconstruction From Monocular Inputs (NeurIPS 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Explicit Geometric Chain-of-Thought for Vision-Language-Action in Autonomous Driving (NeurIPS 2026 · similarity 0.89)
- Uncovering and Shaping the Latent Representation of 3D Scene Topology in Vision-Language Models (NeurIPS 2026 · similarity 0.88)
- Visual Point Cloud Forecasting enables Scalable Autonomous Driving (CVPR 2024 · similarity 0.87)
- Let Language Constrain Geometry: Vision–Language Models as Semantic and Spatial Critics for 3D Generation (ICML 2026 · similarity 0.87)
- Cross-Modal and Uncertainty-Aware Agglomeration for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.87)
- SpaceDrive: Infusing Spatial Awareness into VLM-based Autonomous Driving (CVPR 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
