Research map / 3D vision / 3D generation
SpatialDiff: 3D-Aware Object Movement via Implicit Spatial Modeling
Zheng Liu, Zijian He, Huiguo He, Weizhi Zhong, Yejun Tang, Huan Yang, Kun Gai, Guanbin Li
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 LoRA fine-tuning of DiT attention on the movement dataset (ablation) | Baseline editing model · SpatialBench ablation | GPT-SC: 0.236 -> 0.398 | Improved |
| add LoRA fine-tuning of DiT attention on the movement dataset (ablation) | Baseline editing model · SpatialBench ablation | GPT-PQ: 0.831 -> 0.795 | Hurt |
| add Implicit 3D Spatial Modeling (3D tokens via foundation model) (ablation) | Model A · SpatialBench ablation | GPT-SC: 0.398 -> 0.518 | Improved |
| add Implicit 3D Spatial Modeling (ablation) | Model A · SpatialBench ablation | GPT-PQ: 0.795 -> 0.743 | Hurt |
| add explicit pixel-space depth supervision (ablation) | Model B · SpatialBench ablation | GPT-O: 0.620 -> 0.673 | Improved |
| use latent depth supervision instead of explicit pixel-space depth supervision (ablation) | Model C · SpatialBench ablation | GPT-SC: 0.566 -> 0.804 | Improved |
| use latent depth supervision instead of explicit pixel-space depth supervision (ablation) | Model C · SpatialBench ablation | GPT-PQ: 0.801 -> 0.871 | Improved |
| add Implicit 3D Spatial Modeling (VGGT features) over learnable queries (ablation) | SpatialDiff with LDS · SpatialBench ablation | GPT-SC: 0.656 -> 0.804 | Improved |
| add latent depth supervision to FT+ISM (ablation) | SpatialDiff · SpatialBench ablation | GPT-O: 0.620 -> 0.837 | Improved |
| use SpatialDiff instead of Qwen-Image-Editing | Image editing · SpatialBench | GPT-SC: 0.666 -> 0.803 | Improved |
| use SpatialDiff instead of Nano Banana 2 | Image editing · SpatialBench | Qwen-SC: 0.594 -> 0.778 | Improved |
| use SpatialDiff instead of Nano Banana 2 | Image editing · SpatialBench | GPT-PQ: 0.892 -> 0.886 | No clear effect |
Closest papers by meaning
- Motion4D: Learning 3D-Consistent Motion and Semantics for 4D Scene Understanding (NeurIPS 2025 · similarity 0.86)
- A Diffusion-Based Framework for Occluded Object Movement (AAAI 2025 · similarity 0.86)
- Thinking in Boxes: 3D Editing in Real Images Made Easy (NeurIPS 2026 · similarity 0.86)
- CTRL-D: Controllable Dynamic 3D Scene Editing with Personalized 2D Diffusion (CVPR 2025 · similarity 0.86)
- Move Anything with Layered Scene Diffusion (CVPR 2024 · similarity 0.86)
- MultiDiff: Consistent Novel View Synthesis from a Single Image (CVPR 2024 · similarity 0.85)
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.
