Research map / 3D vision / 3D generation
Counting Stacked Objects
Corentin Dumery, Noa Etté, Aoxiang Fan, Ren Li, Jingyi Xu, Hieu Le, Pascal Fua
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 |
|---|---|---|---|
| use ground-truth depth maps in training instead of estimated ones (ablation) | 3DC · Counting, 100 synthetic scenes | NAE: 0.22 -> 0.28 | Hurt |
| use ground-truth depth maps in both training and evaluation (ablation) | 3DC · Counting, 100 synthetic scenes | NAE: 0.22 -> 0.31 | Hurt |
| use ground-truth depth maps in training instead of estimated ones (ablation) | 3D C occupancy network · Occupancy ratio estimation | R2: 0.79 -> 0.52 | Hurt |
| use ground-truth depth maps in both training and evaluation (ablation) | 3DC occupancy network · Occupancy ratio estimation | R2: 0.79 -> 0.32 | Hurt |
| replace depth map input with RGB image for the gamma-network (ablation) | 3DC · Counting, 45 real scenes | NAE: 0.36 -> 0.57 | Hurt |
| replace depth map input with RGB image for the gamma-network (ablation) | 3DC · Counting, 45 real scenes | sMAPE: 53.31 -> 74.33 | Hurt |
| decompose counting into volume and occupancy ratio vs direct count regression (ViT+H) | 3DC · Counting, 45 real scenes | NAE: 0.94 -> 0.36 | Improved |
| decompose counting vs direct regression (ViT+H) | 3DC · Counting, 100 synthetic scenes | NAE: 0.42 -> 0.22 | Improved |
| learned occupancy network vs depth-ratio baseline with linear correction (DepthCorrected) | occupancy ratio estimation · Occupancy ratio estimation | MAE: 0.10 -> 0.06 | Improved |
| learned network vs mean estimator for occupancy ratio | occupancy ratio estimation · Occupancy ratio estimation | R2: 0.00 -> 0.79 | Improved |
| 3DC vs averaged human guesses (Human-Vote) | 3DC · Counting, 45 real scenes | NAE: 0.60 -> 0.36 | Improved |
| 3DC vs 2D counters evaluated on visible objects only | BMNet+ · Counting, 45 real scenes | sMAPE: 51.28 vs 53.31 | Mixed |
Closest papers by meaning
- From Open Set to Closed Set: Counting Objects by Spatial Divide-and-Conquer (ICCV 2019 · similarity 0.88)
- Wide-Area Crowd Counting: Multi-view Fusion Networks for Counting in Large Scenes (IJCV 2022 · similarity 0.88)
- From Open Set to Closed Set: Supervised Spatial Divide-and-Conquer for Object Counting (IJCV 2023 · similarity 0.88)
- OmniCount: Multi-label Object Counting with Semantic-Geometric Priors (AAAI 2025 · similarity 0.85)
- Learning to Count without Annotations (CVPR 2024 · similarity 0.85)
- DAVE - A Detect-and-Verify Paradigm for Low-Shot Counting (CVPR 2024 · similarity 0.84)
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.
