Research map / 3D vision / 3D generation
RASP: Revisiting 3D Anamorphic Art for Shadow-Guided Packing of Irregular Objects
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes RASP, a differentiable-rendering framework that arranges irregular 3D objects in a container (packing, part assembly, multi-view anamorphic art) by matching shadow/silhouette targets with an SDF-based intersection loss.
Problem
Packing arbitrarily shaped 3D objects into a bounded container with minimal gaps and no intersections, using only 2D shadow or silhouette guidance instead of 3D supervision; also part assembly of broken objects.
Approach
Treat packing as silhouette-matching optimization via differentiable rendering over object rigid poses, and handle mesh intersections and container extrusion with an SDF-based loss that warps precomputed SDFs under rigid transforms instead of costly mesh-mesh tests.
Key results
- RASP achieves an average of 45% occupancy over four IR-BPP categories, versus 34% for Ma et al. (2023).
- Zhao et al. (2021) reaches 51.9% packing efficiency, above RASP's average on the same kind of objects.
- Packing a 106-object Kitchen dataset takes about 17 minutes; Ma et al. average 40.55 minutes versus about 15 minutes for RASP.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use RASP instead of Ma et al. | irregular object packing · Packing comparison | packing efficiency: 34% -> 45% | Helped |
| use RASP instead of Zhao et al. | irregular object packing · online packing dataset comparison | packing efficiency: 51.9% -> 45% | Hurt |
| use RASP instead of Ma et al. | packing optimization · Packing comparison | optimization time: 40.55 minutes -> ∼15 minutes | Helped |
| add object-container extrusion loss Lext (ablation) | RASP · Packing | silhouette loss | Helped |
Surprising findings
- The RL-based online method of Zhao et al. attains higher packing efficiency (51.9%) than RASP's offline optimization (45%).
Limitations
- No physical dynamics such as gravity, so arrangements can float inside the container.
- Struggles with part assembly for more than 2-3 parts, especially symmetric or identical parts.
- Offline method; packing is not physically consistent and is NP-hard so solutions are not optimal.
Experimental setup
Datasets: IR-BPP VOLMAP SilNet Fantastic Breaks
Benchmarks: IR-BPP
Metrics: packing density optimization time silhouette MSE loss
Baselines: Ma et al. (2023) Zhao et al. (IR-BPP RL packing) Wang et al. DBLF HAPE3D
Builds on: Ma et al. 2023 packing optimization Zhao et al. physics-aware online packing
Closest papers by meaning
- Efficient Part-level 3D Object Generation via Dual Volume Packing (NeurIPS 2025 · similarity 0.86)
- Diff3DS: Generating View-Consistent 3D Sketch via Differentiable Curve Rendering (ICLR 2025 · similarity 0.85)
- CASAGPT: Cuboid Arrangement and Scene Assembly for Interior Design (CVPR 2025 · similarity 0.84)
- PartGen: Part-level 3D Generation and Reconstruction with Multi-view Diffusion Models (CVPR 2025 · similarity 0.84)
- MeshArt: Generating Articulated Meshes with Structure-Guided Transformers (CVPR 2025 · similarity 0.84)
- MIDGArD: Modular Interpretable Diffusion over Graphs for Articulated Designs (NeurIPS 2024 · similarity 0.83)
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.