Research map / 3D vision / Face and shape
Affostruction: 3D Affordance Grounding with Generative Reconstruction
Chunghyun Park, Seunghyeon Lee, Minsu Cho
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 depth-conditioned generative reconstruction with sparse voxel fusion | Affostruction · Toys4k reconstruction | IoU: 21.11 -> 32.67 (vs MCC), 19.49 -> 32.67 (vs TRELLIS) | Improved |
| ground affordances with flow-based generation instead of discriminative prediction | Affostruction · Affogato complete geometry | aIoU: 13.6 -> 19.1 | Improved |
| generative flow grounding vs discriminative baselines on AUC, SIM and MAE | Affostruction · Affogato complete geometry | AUC / MAE: AUC 79.0 -> 72.0, MAE 0.111 -> 0.217 | Hurt |
| reconstruct complete geometry before grounding (vs observed-surface grounding) (ablation) | affordance grounding from a single RGBD view · Affogato partial view | aIoU: 0.60 -> 4.74 (MCC+Espresso-3D), 9.26 (Affostruction) | Improved |
| joint sparse-voxel reconstruction and grounding instead of two-stage pipeline | Affostruction · Affogato partial view | aIoU: 4.74 -> 9.26 | Improved |
| add stochastic multi-view training (1-8 views) (ablation) | reconstruction flow transformer · Toys4k reconstruction vs input views | IoU: 33.32 (1 view) -> 46.65 (8 views) | Improved |
| use affordance-driven view selection instead of sequential, 1 extra view (ablation) | Affostruction active view selection · Affogato Furnitures | aIoU: 4.7 -> 9.2 (sequential), 6.2 -> 9.2 (random) | Improved |
| use affordance-driven view selection instead of random at 8 views (ablation) | Affostruction active view selection · Affogato Furnitures | aIoU: 13.2 -> 13.3 | No clear effect |
| fuse features in sparse voxel space, reducing sampling to 5 steps | Affostruction · Affogato runtime | total runtime (s): 8.30 -> 7.20 | Improved |
Closest papers by meaning
- QueryMe: Query-Driven Open-Vocabulary 3D Object Affordances Grounding from Multimodal Evidence (CVPR 2026 · similarity 0.89)
- Grounding 3D Object Affordance with Language Instructions, Visual Observations and Interactions (CVPR 2025 · similarity 0.89)
- Learning 2D Invariant Affordance Knowledge for 3D Affordance Grounding (AAAI 2025 · similarity 0.89)
- UniFunc3D: Unified Active Spatial-Temporal Grounding for 3D Affordance Segmentation (NeurIPS 2026 · similarity 0.88)
- GEAL: Generalizable 3D Affordance Learning with Cross-Modal Consistency (CVPR 2025 · similarity 0.87)
- GREAT: Geometry-Intention Collaborative Inference for Open-Vocabulary 3D Object Affordance Grounding (CVPR 2025 · similarity 0.86)
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.
